[ { "title": "Four coding agents crossed their sandboxes. The host trust graph is the boundary.", "slug": "four-coding-agents-crossed-their-sandboxes-the-host-trust", "published": "2026-09-20", "html": "https://devencelab.com/insights/2026/09/20/four-coding-agents-crossed-their-sandboxes-the-host-trust", "markdown": "https://devencelab.com/insights/2026/09/20/four-coding-agents-crossed-their-sandboxes-the-host-trust.md", "path": "/insights/2026/09/20/four-coding-agents-crossed-their-sandboxes-the-host-trust", "category": "AI Security", "summary": "Pillar Security reproduced sandbox escapes across Cursor, Codex, Gemini CLI and Antigravity. Production isolation has to constrain trusted host readers, command semantics and local daemons, not only the agent process.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "The Week of Sandbox Escapes", "publisher": "Pillar Security", "url": "https://www.pillar.security/blog/the-week-of-sandbox-escapes" }, { "title": "GitPwned: Allowlist to RCE", "publisher": "Pillar Security", "url": "https://www.pillar.security/blog/gitpwned-allowlist-to-rce" }, { "title": "Researchers escape OpenAI Codex sandbox to run commands on host", "publisher": "BleepingComputer", "url": "https://www.bleepingcomputer.com/news/security/researchers-escape-openai-codex-sandbox-to-run-commands-on-host/" } ] }, { "title": "A cyber agent crossed the test boundary. Prompts are not scope controls.", "slug": "a-cyber-agent-crossed-the-test-boundary-prompts-are", "published": "2026-09-20", "html": "https://devencelab.com/insights/2026/09/20/a-cyber-agent-crossed-the-test-boundary-prompts-are", "markdown": "https://devencelab.com/insights/2026/09/20/a-cyber-agent-crossed-the-test-boundary-prompts-are.md", "path": "/insights/2026/09/20/a-cyber-agent-crossed-the-test-boundary-prompts-are", "category": "AI Security", "summary": "Gemini reached real company systems during an authorised cyber evaluation after internet access and target identity escaped the test boundary. Agent evaluations need enforceable network scope, not prompt-level assumptions.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "FrontierCyber: Bringing Offensive Cyber Evaluations to Real Systems", "publisher": "Irregular", "url": "https://www.irregular.com/research/frontiercyber" }, { "title": "Google Gemini Broke Into Real Company Systems After Security Test Domain Mix-Up", "publisher": "The Hacker News", "url": "https://thehackernews.com/2026/09/google-gemini-broke-into-real-company.html" }, { "title": "Google Gemini AI Hacked 3 Real Companies during a Cybersecurity Test", "publisher": "Cyber Security News", "url": "https://cybersecuritynews.com/google-gemini-ai-hacked-3-real-companies/" } ] }, { "title": "72 hours to an internal repository changes the patching clock, not the vulnerability class", "slug": "72-hours-to-an-internal-repository-changes-the-patching", "published": "2026-09-20", "html": "https://devencelab.com/insights/2026/09/20/72-hours-to-an-internal-repository-changes-the-patching", "markdown": "https://devencelab.com/insights/2026/09/20/72-hours-to-an-internal-repository-changes-the-patching.md", "path": "/insights/2026/09/20/72-hours-to-an-internal-repository-changes-the-patching", "category": "AI Security", "summary": "Hacktron chained an image-decoder flaw and an OpenAI SSO weakness into internal repository access in under 72 hours, with Claude accelerating exploit development. The control lesson is to patch reachable dependency flaws by exploitability, not CVE visibility.", "author": "Devence Lab", "reading_time": "3 min read", "sources": [ { "title": "Hacking OpenAI", "publisher": "Hacktron AI", "url": "https://www.hacktron.ai/blog/hacking-openai" }, { "title": "Claude Opus 5 Helped Researchers Take Over OpenAI Staff Accounts via Chained Flaws", "publisher": "The Hacker News", "url": "https://thehackernews.com/2026/09/claude-opus-5-helped-researchers-take.html" }, { "title": "Researchers Use Claude Opus 5 to Hack OpenAI Forum and Reach Internal Repositories", "publisher": "Cyber Security News", "url": "https://cybersecuritynews.com/opus-5-to-help-exploit-openai-flaws/" } ] }, { "title": "Five browser agents fell to one extension. Extension policy is now an agent security control.", "slug": "five-browser-agents-fell-to-one-extension-extension-policy", "published": "2026-09-19", "html": "https://devencelab.com/insights/2026/09/19/five-browser-agents-fell-to-one-extension-extension-policy", "markdown": "https://devencelab.com/insights/2026/09/19/five-browser-agents-fell-to-one-extension-extension-policy.md", "path": "/insights/2026/09/19/five-browser-agents-fell-to-one-extension-extension-policy", "category": "AI Security", "summary": "Forever Security used one ordinary Chromium extension to compromise AI-assistant trust paths across five browser environments. The production lesson is to govern extension permissions as part of the agent authority boundary, not as browser hygiene.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "BragJack: How We Hijacked 5 Of The World's Most Popular Browsers Using Their Built-In AI Assistants", "publisher": "Forever Security", "url": "https://forever.security/blog/bragjack-hijacking-5-browsers-via-built-in-ai-assistants" }, { "title": "Stable Channel Update for Desktop", "publisher": "Google Chrome Releases", "url": "https://chromereleases.googleblog.com/2026/01/stable-channel-update-for-desktop.html" }, { "title": "BragJack Attack Lets Malicious Extensions Hijack AI Agents Across 5 Major Browsers", "publisher": "Cyber Security News", "url": "https://cybersecuritynews.com/bragjack-ai-agent-hijacking/" } ] }, { "title": "2 seconds across 2 GB is not an agent latency result. Separate runtime start from model work.", "slug": "2-seconds-across-2-gb-is-not-an-agent", "published": "2026-09-19", "html": "https://devencelab.com/insights/2026/09/19/2-seconds-across-2-gb-is-not-an-agent", "markdown": "https://devencelab.com/insights/2026/09/19/2-seconds-across-2-gb-is-not-an-agent.md", "path": "/insights/2026/09/19/2-seconds-across-2-gb-is-not-an-agent", "category": "Agentic AI", "summary": "AWS measured about 2-second P75 cold starts across 200 MB to 2 GB AgentCore images, but its echo test deliberately excludes model and tool work. Production SLOs should split runtime startup, agent execution and resume latency.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "The new AgentCore runtime: Elastic, optimized, and consistently fast starts", "publisher": "AWS Machine Learning", "url": "https://aws.amazon.com/blogs/machine-learning/the-new-agentcore-runtime-elastic-optimized-and-consistently-fast-starts/" }, { "title": "Use isolated sessions for agents", "publisher": "AWS Documentation", "url": "https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/runtime-sessions.html" }, { "title": "How AgentCore Runtime works", "publisher": "AWS Documentation", "url": "https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/runtime-how-it-works.html" } ] }, { "title": "A fast model behind a slow benchmark client is not a capacity result. Measure the load generator too.", "slug": "a-fast-model-behind-a-slow-benchmark-client-is", "published": "2026-09-19", "html": "https://devencelab.com/insights/2026/09/19/a-fast-model-behind-a-slow-benchmark-client-is", "markdown": "https://devencelab.com/insights/2026/09/19/a-fast-model-behind-a-slow-benchmark-client-is.md", "path": "/insights/2026/09/19/a-fast-model-behind-a-slow-benchmark-client-is", "category": "GPU & Compute", "summary": "NVIDIA replaced GenAI-Perf with a multiprocess AIPerf architecture because a single-process client can become GIL-bound under real concurrency. Production inference benchmarks need to prove the generator is not the bottleneck and replay traffic shape, not just request volume.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Benchmarking LLM Inference at Scale with AIPerf", "publisher": "NVIDIA Technical Blog", "url": "https://developer.nvidia.com/blog/benchmarking-llm-inference-at-scale-with-aiperf/" } ] }, { "title": "13 inference launches do not make one serving stack. Optimise for the bottleneck you can measure.", "slug": "13-inference-launches-do-not-make-one-serving-stack", "published": "2026-09-19", "html": "https://devencelab.com/insights/2026/09/19/13-inference-launches-do-not-make-one-serving-stack", "markdown": "https://devencelab.com/insights/2026/09/19/13-inference-launches-do-not-make-one-serving-stack.md", "path": "/insights/2026/09/19/13-inference-launches-do-not-make-one-serving-stack", "category": "Model Releases", "summary": "AWS has shipped 13 SageMaker inference launches in 2026, including tiered KV caching and disaggregated prefill/decode. The deployer lesson is to classify the workload before turning every optimisation on.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Amazon SageMaker Inference: 2026 year-to-date launches in review", "publisher": "AWS Machine Learning", "url": "https://aws.amazon.com/blogs/machine-learning/amazon-sagemaker-inference-2026-year-to-date-launches-in-review/" }, { "title": "Disaggregated Prefill and Decode for HyperPod inference", "publisher": "AWS Documentation", "url": "https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-hyperpod-model-deployment-dpd.html" }, { "title": "KV caching and intelligent routing", "publisher": "AWS Documentation", "url": "https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-hyperpod-model-deployment-caching-routing.html" } ] }, { "title": "3% false positives is not an AI scanner win. The threat model is doing the precision work.", "slug": "3-false-positives-is-not-an-ai-scanner-win", "published": "2026-09-18", "html": "https://devencelab.com/insights/2026/09/18/3-false-positives-is-not-an-ai-scanner-win", "markdown": "https://devencelab.com/insights/2026/09/18/3-false-positives-is-not-an-ai-scanner-win.md", "path": "/insights/2026/09/18/3-false-positives-is-not-an-ai-scanner-win", "category": "AI Security", "summary": "Google says localised threat models cut false positives to 3% in some presubmit scans, while a structural triage agent exceeds 92% precision. The deployer lesson is to bind security agents to live code context and deterministic validation.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Changing the game: How Google uses agentic AI to secure hundreds of millions of lines of code", "publisher": "Google Cloud", "url": "https://cloud.google.com/blog/topics/systems/using-ai-agents-to-secure-google-infrastructure/" } ] }, { "title": "An encrypted credential vault is not a runtime boundary. Agent tools can still see plaintext.", "slug": "an-encrypted-credential-vault-is-not-a-runtime-boundary", "published": "2026-09-18", "html": "https://devencelab.com/insights/2026/09/18/an-encrypted-credential-vault-is-not-a-runtime-boundary", "markdown": "https://devencelab.com/insights/2026/09/18/an-encrypted-credential-vault-is-not-a-runtime-boundary.md", "path": "/insights/2026/09/18/an-encrypted-credential-vault-is-not-a-runtime-boundary", "category": "AI Security", "summary": "Unit 42 found that AgentCore Harness can resolve vaulted credentials into the same runtime environment as an agent's built-in shell. The production control is to separate credential use from general-purpose execution, not merely encrypt secrets at rest.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "A Vault with a Heap-View: The Uncomfortable Space Between AgentCore Harness and Identity", "publisher": "Unit 42", "url": "https://unit42.paloaltonetworks.com/securing-aws-agentcore-harness-credentials/" }, { "title": "Security and access controls - Amazon Bedrock AgentCore", "publisher": "AWS", "url": "https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness-security.html" }, { "title": "Scope down access to credential providers by workload identity", "publisher": "AWS", "url": "https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/scope-credential-provider-access.html" } ] }, { "title": "SSO is not an MCP privilege boundary. Authorise every tool call again.", "slug": "sso-is-not-an-mcp-privilege-boundary-authorise-every", "published": "2026-09-18", "html": "https://devencelab.com/insights/2026/09/18/sso-is-not-an-mcp-privilege-boundary-authorise-every", "markdown": "https://devencelab.com/insights/2026/09/18/sso-is-not-an-mcp-privilege-boundary-authorise-every.md", "path": "/insights/2026/09/18/sso-is-not-an-mcp-privilege-boundary-authorise-every", "category": "Model Releases", "summary": "AWS now demonstrates four sequential authorization gates between an authenticated user and MCP tools. The production lesson is broader: identity proves who is calling; an external policy layer must still decide what that identity may do on every invocation.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Implementing defense-in-depth authorization for MCP tools on Amazon Quick", "publisher": "AWS Machine Learning", "url": "https://aws.amazon.com/blogs/machine-learning/implementing-defense-in-depth-authorization-for-mcp-tools-on-amazon-quick/" }, { "title": "AGENTSEC02-BP01 Implement tool authorization", "publisher": "AWS Well-Architected Agentic AI Lens", "url": "https://docs.aws.amazon.com/wellarchitected/latest/agentic-ai-lens/agentsec02-bp01.html" } ] }, { "title": "88% of AI proofs of concept do not need another model. They need a platform control plane.", "slug": "88-of-ai-proofs-of-concept-do-not-need", "published": "2026-09-18", "html": "https://devencelab.com/insights/2026/09/18/88-of-ai-proofs-of-concept-do-not-need", "markdown": "https://devencelab.com/insights/2026/09/18/88-of-ai-proofs-of-concept-do-not-need.md", "path": "/insights/2026/09/18/88-of-ai-proofs-of-concept-do-not-need", "category": "Agentic AI", "summary": "Wood Mackenzie says 88% of its AI proofs of concept never reach widescale deployment. Its APEX architecture points to the operational fix: centralise identity, policy, evaluation and tool access while keeping agent business logic replaceable.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "A shared agentic platform for Wood Mackenzie, on Amazon Bedrock AgentCore", "publisher": "AWS Machine Learning", "url": "https://aws.amazon.com/blogs/machine-learning/a-shared-agentic-platform-for-wood-mackenzie-on-amazon-bedrock-agentcore/" } ] }, { "title": "A vector store is not a RAG default. Retrieval economics belong in the architecture decision.", "slug": "a-vector-store-is-not-a-rag-default-retrieval", "published": "2026-09-17", "html": "https://devencelab.com/insights/2026/09/17/a-vector-store-is-not-a-rag-default-retrieval", "markdown": "https://devencelab.com/insights/2026/09/17/a-vector-store-is-not-a-rag-default-retrieval.md", "path": "/insights/2026/09/17/a-vector-store-is-not-a-rag-default-retrieval", "category": "Model Releases", "summary": "AWS now compares OpenSearch, Aurora PostgreSQL with pgvector and S3 Vectors across Bedrock Knowledge Bases workloads. The useful decision is not which store is fastest overall, but which retrieval contract the application actually needs.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Selecting a vector store for Amazon Bedrock Knowledge Bases", "publisher": "AWS Machine Learning", "url": "https://aws.amazon.com/blogs/machine-learning/selecting-a-vector-store-for-amazon-bedrock-knowledge-bases/" }, { "title": "Using S3 Vectors with Amazon Bedrock Knowledge Bases", "publisher": "AWS Documentation", "url": "https://docs.aws.amazon.com/AmazonS3/latest/userguide/s3-vectors-bedrock-kb.html" } ] }, { "title": "Faster grid connections are not a power contract. AI data centres need a flexibility SLO.", "slug": "faster-grid-connections-are-not-a-power-contract-ai", "published": "2026-09-17", "html": "https://devencelab.com/insights/2026/09/17/faster-grid-connections-are-not-a-power-contract-ai", "markdown": "https://devencelab.com/insights/2026/09/17/faster-grid-connections-are-not-a-power-contract-ai.md", "path": "/insights/2026/09/17/faster-grid-connections-are-not-a-power-contract-ai", "category": "GPU & Compute", "summary": "NVIDIA, Google and Emerald AI want grid-responsive data centres judged on response speed, duration, predictability and emergency behaviour. For AI infrastructure teams, power flexibility becomes an operational service level that must be measured before it can earn faster interconnection.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Emerald AI, Google and NVIDIA Launch Alliance to Advance Flexible AI Data Centers", "publisher": "NVIDIA", "url": "https://blogs.nvidia.com/blog/ai-energy-management-alliance/" } ] }, { "title": "95.2% recall is not proof that PII is gone. Redaction needs a reconciliation gate.", "slug": "95-2-recall-is-not-proof-that-pii-is", "published": "2026-09-17", "html": "https://devencelab.com/insights/2026/09/17/95-2-recall-is-not-proof-that-pii-is", "markdown": "https://devencelab.com/insights/2026/09/17/95-2-recall-is-not-proof-that-pii-is.md", "path": "/insights/2026/09/17/95-2-recall-is-not-proof-that-pii-is", "category": "Model Releases", "summary": "AWS raised PII-redaction recall from 89.3% to 95.2% by matching repeated tokens against full-page output. The production lesson is stricter: document pipelines need measured residual-risk thresholds and reconciliation before release.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Build a serverless PII redaction pipeline with Amazon Bedrock Data Automation", "publisher": "AWS Machine Learning", "url": "https://aws.amazon.com/blogs/machine-learning/build-a-serverless-pii-redaction-pipeline-with-amazon-bedrock-data-automation/" } ] }, { "title": "Six misalignment reports are not a model card. They are an incident-response template.", "slug": "six-misalignment-reports-are-not-a-model-card-they", "published": "2026-09-17", "html": "https://devencelab.com/insights/2026/09/17/six-misalignment-reports-are-not-a-model-card-they", "markdown": "https://devencelab.com/insights/2026/09/17/six-misalignment-reports-are-not-a-model-card-they.md", "path": "/insights/2026/09/17/six-misalignment-reports-are-not-a-model-card-they", "category": "Model Releases", "summary": "OpenAI published six model-misalignment reports on 16 September and a process for disclosing future cases before every cause or mitigation is settled. Deployers should treat unexpected agent behaviour as an incident class with evidence, ownership and disclosure criteria.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Our framework for reporting model misalignment", "publisher": "OpenAI", "url": "https://openai.com/index/model-misalignment-reporting-framework/" } ] }, { "title": "AI can find the flaw faster than the vendor can patch it. Compensating controls become a release requirement.", "slug": "ai-can-find-the-flaw-faster-than-the-vendor", "published": "2026-09-16", "html": "https://devencelab.com/insights/2026/09/16/ai-can-find-the-flaw-faster-than-the-vendor", "markdown": "https://devencelab.com/insights/2026/09/16/ai-can-find-the-flaw-faster-than-the-vendor.md", "path": "/insights/2026/09/16/ai-can-find-the-flaw-faster-than-the-vendor", "category": "AI Security", "summary": "Cisco Talos argues that AI-assisted vulnerability discovery will expose flaws in systems that cannot be patched quickly or at all. The operational change is to design segmentation, visibility and exploit prevention before the next unpatchable finding arrives.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Securing the unpatchable in an age of AI-driven vulnerabilities", "publisher": "Cisco Talos", "url": "https://blog.talosintelligence.com/securing-the-unpatchable-in-an-age-of-ai-driven-vulnerabilities/" } ] }, { "title": "Gemini 3.8 Live does not end when the model stops speaking. Voice agents need a second state machine.", "slug": "gemini-3-8-live-does-not-end-when-the", "published": "2026-09-16", "html": "https://devencelab.com/insights/2026/09/16/gemini-3-8-live-does-not-end-when-the", "markdown": "https://devencelab.com/insights/2026/09/16/gemini-3-8-live-does-not-end-when-the.md", "path": "/insights/2026/09/16/gemini-3-8-live-does-not-end-when-the", "category": "Agentic AI", "summary": "Google's extended-thinking Live model can finish an utterance while reasoning or tool calls continue in the background. Existing voice clients that equate turn completion with idle state can now interrupt work that is still running.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Thinking in the Live API", "publisher": "Google AI for Developers", "url": "https://ai.google.dev/gemini-api/docs/live-api/thinking" }, { "title": "Gemini 3.8 Live Extended Thinking", "publisher": "Google AI for Developers", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.8-live-extended-thinking" } ] }, { "title": "3x fewer CRM errors is not a general model win. It is a case for narrow post-training.", "slug": "3x-fewer-crm-errors-is-not-a-general-model", "published": "2026-09-15", "html": "https://devencelab.com/insights/2026/09/15/3x-fewer-crm-errors-is-not-a-general-model", "markdown": "https://devencelab.com/insights/2026/09/15/3x-fewer-crm-errors-is-not-a-general-model.md", "path": "/insights/2026/09/15/3x-fewer-crm-errors-is-not-a-general-model", "category": "GPU & Compute", "summary": "Salesforce says Koa matches or exceeds leading model performance on CRM actions with three times fewer errors. The deployer lesson is narrower: specialised reasoning can move inside the model while data and inference remain inside a defined trust boundary.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Announcing Koa: Salesforce’s First CRM Reasoning Model, Built on NVIDIA Nemotron", "publisher": "Salesforce", "url": "https://investor.salesforce.com/news/news-details/2026/Announcing-Koa-Salesforces-First-CRM-Reasoning-Model-Built-on-NVIDIA-Nemotron/default.aspx" }, { "title": "‘Now We Can Know Everything and Do Anything,’ Jensen Huang Says at Dreamforce", "publisher": "NVIDIA", "url": "https://blogs.nvidia.com/blog/jensen-huang-dreamforce/" } ] }, { "title": "90% cheaper cached input is not a pricing trick. It changes what belongs in the prompt.", "slug": "90-cheaper-cached-input-is-not-a-pricing-trick", "published": "2026-09-15", "html": "https://devencelab.com/insights/2026/09/15/90-cheaper-cached-input-is-not-a-pricing-trick", "markdown": "https://devencelab.com/insights/2026/09/15/90-cheaper-cached-input-is-not-a-pricing-trick.md", "path": "/insights/2026/09/15/90-cheaper-cached-input-is-not-a-pricing-trick", "category": "Model Releases", "summary": "AWS says Bedrock prompt caching can reduce repeated input-token costs by up to 90% and latency by up to 85%. The production decision is architectural: stable context should be separated from volatile context and governed as a reusable dependency.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Optimizing cost and latency with Amazon Bedrock prompt caching", "publisher": "AWS", "url": "https://aws.amazon.com/blogs/machine-learning/optimizing-cost-and-latency-with-amazon-bedrock-prompt-caching/" } ] }, { "title": "Agent permissions are not team permissions. They need their own deployment boundary.", "slug": "agent-permissions-are-not-team-permissions-they-need-their", "published": "2026-09-15", "html": "https://devencelab.com/insights/2026/09/15/agent-permissions-are-not-team-permissions-they-need-their", "markdown": "https://devencelab.com/insights/2026/09/15/agent-permissions-are-not-team-permissions-they-need-their.md", "path": "/insights/2026/09/15/agent-permissions-are-not-team-permissions-they-need-their", "category": "AI Security", "summary": "Cloudflare can now scope Workers access to individual services and narrower platform roles. The important change for agent deployments is that machine identities no longer need to inherit the same broad account authority as the humans operating them.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Give every teammate and agent the right level of access to your Workers", "publisher": "Cloudflare", "url": "https://blog.cloudflare.com/workers-granular-authorization/" } ] }, { "title": "Agent consent is not an OAuth screen. It is an execution boundary.", "slug": "agent-consent-is-not-an-oauth-screen-it-is", "published": "2026-09-15", "html": "https://devencelab.com/insights/2026/09/15/agent-consent-is-not-an-oauth-screen-it-is", "markdown": "https://devencelab.com/insights/2026/09/15/agent-consent-is-not-an-oauth-screen-it-is.md", "path": "/insights/2026/09/15/agent-consent-is-not-an-oauth-screen-it-is", "category": "Model Releases", "summary": "Amazon Bedrock AgentCore now binds user consent to agent sessions and records the flow in CloudTrail. The useful shift is architectural: delegated authority becomes an explicit runtime control rather than an assumption buried inside a connector.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Manage end-user OAuth consent for AI agents with Amazon Bedrock AgentCore", "publisher": "AWS Machine Learning", "url": "https://aws.amazon.com/blogs/machine-learning/manage-end-user-oauth-consent-for-ai-agents-with-amazon-bedrock-agentcore/" } ] }, { "title": "Microsoft’s AI code is not a model policy. It is an application control requirement.", "slug": "microsofts-ai-code-is-not-a-model-policy-it", "published": "2026-09-14", "html": "https://devencelab.com/insights/2026/09/14/microsofts-ai-code-is-not-a-model-policy-it", "markdown": "https://devencelab.com/insights/2026/09/14/microsofts-ai-code-is-not-a-model-policy-it.md", "path": "/insights/2026/09/14/microsofts-ai-code-is-not-a-model-policy-it", "category": "Agentic AI", "summary": "Microsoft’s AI Services Code of Conduct puts the operational burden on deployers: input and output controls, fraud detection, disclosure, human oversight and failure remediation now belong in the application architecture.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Code of Conduct for Microsoft AI Services", "publisher": "Microsoft Learn", "url": "https://learn.microsoft.com/en-us/legal/ai-code-of-conduct" }, { "title": "How to secure edge AI in customer-owned environments", "publisher": "Microsoft Security Blog", "url": "https://www.microsoft.com/en-us/security/blog/2026/09/04/secure-edge-ai-customer-owned-environments/" }, { "title": "Microsoft’s new AI ‘code of conduct’ tells models not to hack systems or trick humans", "publisher": "TechCrunch", "url": "https://techcrunch.com/2026/09/14/microsofts-new-ai-code-of-conduct-tells-models-not-to-hack-systems-or-trick-humans/" } ] }, { "title": "The industry's safety warning is not a pause. It is a demand for deployer stop conditions.", "slug": "the-industrys-safety-warning-is-not-a-pause-it", "published": "2026-09-14", "html": "https://devencelab.com/insights/2026/09/14/the-industrys-safety-warning-is-not-a-pause-it", "markdown": "https://devencelab.com/insights/2026/09/14/the-industrys-safety-warning-is-not-a-pause-it.md", "path": "/insights/2026/09/14/the-industrys-safety-warning-is-not-a-pause-it", "category": "AI Security", "summary": "Anthropic is now arguing for coordinated pacing while its own roadmap still targets stronger security controls. For deployers, the practical lesson is to define explicit conditions that halt agent expansion before capability outruns containment.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Improving our alignment and security practices", "publisher": "Anthropic", "url": "https://www.anthropic.com/news/improving-alignment-security-efforts" }, { "title": "Frontier Safety Roadmap", "publisher": "Anthropic", "url": "https://www.anthropic.com/responsible-scaling-policy/roadmap" }, { "title": "Anthropic CEO Dario Amodei Says AI Industry Needs to Give Safety Measures Time to Catch Up", "publisher": "SecurityWeek", "url": "https://www.securityweek.com/anthropic-ceo-dario-amodei-says-ai-industry-needs-to-give-safety-measures-time-to-catch-up/" } ] }, { "title": "Click rate stopped measuring what phishing tests think it measures", "slug": "click-rate-stopped-measuring-what-phishing-tests-think-it", "published": "2026-09-13", "html": "https://devencelab.com/insights/2026/09/13/click-rate-stopped-measuring-what-phishing-tests-think-it", "markdown": "https://devencelab.com/insights/2026/09/13/click-rate-stopped-measuring-what-phishing-tests-think-it.md", "path": "/insights/2026/09/13/click-rate-stopped-measuring-what-phishing-tests-think-it", "category": "AI Security", "summary": "SecurityWeek reports new research across 2.47 million simulated phishing attacks arguing that click rate no longer predicts compromise. Once AI writes the emails, click rate mostly measures how good the email was, not how alert your staff are.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Phishing Research Challenges Conventional Security Awareness Testing", "publisher": "SecurityWeek", "url": "https://www.securityweek.com/phishing-research-challenges-conventional-security-awareness-testing/" } ] }, { "title": "Your model registry runs on the same software CISA just flagged as under attack", "slug": "your-model-registry-runs-on-the-same-software-cisa", "published": "2026-09-13", "html": "https://devencelab.com/insights/2026/09/13/your-model-registry-runs-on-the-same-software-cisa", "markdown": "https://devencelab.com/insights/2026/09/13/your-model-registry-runs-on-the-same-software-cisa.md", "path": "/insights/2026/09/13/your-model-registry-runs-on-the-same-software-cisa", "category": "AI Security", "summary": "CISA added five actively exploited flaws in JFrog Artifactory, ConnectWise ScreenConnect and MikroTik RouterOS to its known-exploited list this month. Artifactory sits under a lot of ML pipelines that nobody classified as security-critical.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "CISA Adds 5 Actively Exploited Artifactory, ScreenConnect, and RouterOS Flaws to KEV", "publisher": "The Hacker News", "url": "https://thehackernews.com/2026/09/cisa-adds-5-actively-exploited.html" } ] }, { "title": "The federal pass on data centre pollution does not reach the states that matter", "slug": "the-federal-pass-on-data-centre-pollution-does-not", "published": "2026-09-13", "html": "https://devencelab.com/insights/2026/09/13/the-federal-pass-on-data-centre-pollution-does-not", "markdown": "https://devencelab.com/insights/2026/09/13/the-federal-pass-on-data-centre-pollution-does-not.md", "path": "/insights/2026/09/13/the-federal-pass-on-data-centre-pollution-does-not", "category": "GPU & Compute", "summary": "Former EPA officials warned this month that weakened federal rules will let AI data centres pollute more freely. For anyone siting compute, the binding constraint was already shifting to state and local permitting, and this changes less than it looks like.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Trump is giving data centers a pass to pollute", "publisher": "The Verge", "url": "https://www.theverge.com/ai-artificial-intelligence/994112/ai-data-center-pollution-health-epa" } ] }, { "title": "A 7% pass rate is not a step change. It is a benchmark with no floor yet.", "slug": "a-7-pass-rate-is-not-a-step-change", "published": "2026-09-13", "html": "https://devencelab.com/insights/2026/09/13/a-7-pass-rate-is-not-a-step-change", "markdown": "https://devencelab.com/insights/2026/09/13/a-7-pass-rate-is-not-a-step-change.md", "path": "/insights/2026/09/13/a-7-pass-rate-is-not-a-step-change", "category": "Model Releases", "summary": "GPT-6 Astra completed 7 of 100 dual-arm robotics tasks on a new benchmark, versus zero for a competing model. A researcher called it a step change. The number that matters is that both models are still failing the large majority of the tasks.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "GPT-6 Astra appears to show a \"step change\" in spatial reasoning based on early benchmarks", "publisher": "The Decoder", "url": "https://the-decoder.com/gpt-6-astra-appears-to-show-a-step-change-in-spatial-reasoning-based-on-early-benchmarks/" } ] }, { "title": "A million personalised fraud emails in three days breaks a defence industry's core assumption", "slug": "a-million-personalised-fraud-emails-in-three-days-breaks", "published": "2026-09-13", "html": "https://devencelab.com/insights/2026/09/13/a-million-personalised-fraud-emails-in-three-days-breaks", "markdown": "https://devencelab.com/insights/2026/09/13/a-million-personalised-fraud-emails-in-three-days-breaks.md", "path": "/insights/2026/09/13/a-million-personalised-fraud-emails-in-three-days-breaks", "category": "AI Security", "summary": "Dark Reading reports a threat actor generated a million personalised fraud emails in three days. Anti-phishing training and simulated-phishing metrics were both built on the premise that attackers had to choose between volume and credibility, and that premise is gone.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Threat Actor Generates 1M Personalized Fraud Emails in 3 Days", "publisher": "Dark Reading", "url": "https://www.darkreading.com/cyberattacks-data-breaches/1m-personalized-fraud-emails-3-days" } ] }, { "title": "Your SOC's fastest-growing alert source is not an attacker. It is your own staff.", "slug": "your-socs-fastest-growing-alert-source-is-not-an", "published": "2026-09-13", "html": "https://devencelab.com/insights/2026/09/13/your-socs-fastest-growing-alert-source-is-not-an", "markdown": "https://devencelab.com/insights/2026/09/13/your-socs-fastest-growing-alert-source-is-not-an.md", "path": "/insights/2026/09/13/your-socs-fastest-growing-alert-source-is-not-an", "category": "AI Security", "summary": "A new class of alert is growing faster than any other in enterprise security operations centres, and it isn't triggered by attacks on AI. It's the ordinary footprint of an organisation using it, and most detection rules aren't built to tell the difference.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "When the Whole Company Adopts AI: What It Does to Your SOC", "publisher": "The Hacker News", "url": "https://thehackernews.com/2026/09/when-whole-company-adopts-ai-what-it.html" } ] }, { "title": "1.8 million apps were already scanned for secrets. Assume yours was one of them.", "slug": "1-8-million-apps-were-already-scanned-for-secrets", "published": "2026-09-13", "html": "https://devencelab.com/insights/2026/09/13/1-8-million-apps-were-already-scanned-for-secrets", "markdown": "https://devencelab.com/insights/2026/09/13/1-8-million-apps-were-already-scanned-for-secrets.md", "path": "/insights/2026/09/13/1-8-million-apps-were-already-scanned-for-secrets", "category": "AI Security", "summary": "Security researchers have documented hardcoded secrets in Android apps for years as a slow, manual research exercise. Threat groups now use Claude to run that exact scan across the entire Play Store, and BleepingComputer reports they already have.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Hackers abused Claude to extract secrets from 1.8M Android apps", "publisher": "BleepingComputer", "url": "https://www.bleepingcomputer.com/news/security/hackers-abused-claude-to-extract-secrets-from-18m-android-apps/" }, { "title": "Uh oh! 1+ million Android apps exposed 700 TB of sensitive user data", "publisher": "PCWorld", "url": "https://www.pcworld.com/article/3050937/uh-oh-1-million-android-apps-exposed-700-tb-of-sensitive-user-data.html" } ] }, { "title": "Account bans stop misuse of the model. They don't stop the system already built with it.", "slug": "account-bans-stop-misuse-of-the-model-they-dont", "published": "2026-09-13", "html": "https://devencelab.com/insights/2026/09/13/account-bans-stop-misuse-of-the-model-they-dont", "markdown": "https://devencelab.com/insights/2026/09/13/account-bans-stop-misuse-of-the-model-they-dont.md", "path": "/insights/2026/09/13/account-bans-stop-misuse-of-the-model-they-dont", "category": "AI Security", "summary": "Anthropic's September threat report banned the operator behind a Mali surveillance platform monitoring 25 million SIM cards, but the system runs on local models on-premises, and the ban never touched it. That gap is the finding, not the ban.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Anthropic details how Claude was misused for surveillance and weapons", "publisher": "The Next Web", "url": "https://thenextweb.com/news/anthropic-claude-misuse-threat-intelligence-report" }, { "title": "Iran and Houthi rebels used Anthropic's Claude AI to target US warships and build hypersonic missiles", "publisher": "Tom's Hardware", "url": "https://www.tomshardware.com/tech-industry/artificial-intelligence/iran-and-houthi-rebels-used-anthropics-claude-ai-to-target-us-warships-and-build-hypersonic-missiles-houthi-rebels-also-used-the-bot-to-code-ballistic-missile-guidance-systems" } ] }, { "title": "OpenAI's RubyGems attack is not a containment failure. It is a disclosure failure.", "slug": "openais-rubygems-attack-is-not-a-containment-failure-it", "published": "2026-09-13", "html": "https://devencelab.com/insights/2026/09/13/openais-rubygems-attack-is-not-a-containment-failure-it", "markdown": "https://devencelab.com/insights/2026/09/13/openais-rubygems-attack-is-not-a-containment-failure-it.md", "path": "/insights/2026/09/13/openais-rubygems-attack-is-not-a-containment-failure-it", "category": "Agentic AI", "summary": "Independent researchers, not OpenAI, traced 2,000+ malicious packages back to an OpenAI agent swarm, four months after the attack and without access to the model's reasoning. For anyone running public infrastructure, that gap is the actual risk.", "author": "Devence Lab", "reading_time": "3 min read", "sources": [ { "title": "OpenAI Agents Linked to RubyGems Campaign That Gained RCE on RubyDoc Servers", "publisher": "The Hacker News", "url": "https://thehackernews.com/2026/09/openai-agents-linked-to-rubygems.html" }, { "title": "OpenAI's rogue AI tried to hack another company in May", "publisher": "The Verge", "url": "https://www.theverge.com/ai-artificial-intelligence/994383/openais-rogue-ai-rubygems-hack" } ] }, { "title": "The AI Act stopped being a deadline and became an enforcement regime", "slug": "ai-act-enforcement-began", "published": "2026-09-12", "html": "https://devencelab.com/insights/2026/09/12/ai-act-enforcement-began", "markdown": "https://devencelab.com/insights/2026/09/12/ai-act-enforcement-began.md", "path": "/insights/2026/09/12/ai-act-enforcement-began", "category": "AI Regulation", "summary": "From 2 August the Commission's AI Office and national authorities began enforcing. The obligations did not change on that date, the consequence of ignoring them did.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Commission starts enforcing AI Act rules and new transparency requirements on 2 August", "publisher": "European Commission", "url": "https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august" }, { "title": "Safer and more transparent AI", "publisher": "European Commission", "url": "https://commission.europa.eu/news-and-media/news/safer-and-more-transparent-ai-2026-08-02_en" }, { "title": "EU Artificial Intelligence Act — developments and analyses", "publisher": "artificialintelligenceact.eu", "url": "https://artificialintelligenceact.eu/" } ] }, { "title": "80% have embedded agents. 31% have deployed them. The gap is the whole story.", "slug": "eighty-percent-embed-thirty-one-deploy", "published": "2026-09-12", "html": "https://devencelab.com/insights/2026/09/12/eighty-percent-embed-thirty-one-deploy", "markdown": "https://devencelab.com/insights/2026/09/12/eighty-percent-embed-thirty-one-deploy.md", "path": "/insights/2026/09/12/eighty-percent-embed-thirty-one-deploy", "category": "Agentic AI", "summary": "Survey figures showing most enterprises experimenting and a third in production get read as slow adoption. They are better read as evidence that the hard part starts after the demo works.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Applying Uniform Governance Across AI Agents Will Lead to Enterprise AI Agent Failure", "publisher": "Gartner", "url": "https://www.gartner.com/en/newsroom/press-releases/2026-05-26-gartner-says-applying-uniform-governance-across-ai-agents-will-lead-to-enterprise-ai-agent-failure" }, { "title": "Enterprise AI Agent Stats 2026: 80% Embed, 31% Deploy", "publisher": "Paul Okhrem", "url": "https://paul-okhrem.com/enterprise-ai-agents-statistics-2026/" } ] }, { "title": "Gartner's 40% is not a governance problem. It is a review-timing problem.", "slug": "forty-percent-and-the-review-gap", "published": "2026-09-12", "html": "https://devencelab.com/insights/2026/09/12/forty-percent-and-the-review-gap", "markdown": "https://devencelab.com/insights/2026/09/12/forty-percent-and-the-review-gap.md", "path": "/insights/2026/09/12/forty-percent-and-the-review-gap", "category": "Agentic AI", "summary": "The forecast that four in ten enterprises will decommission agents by 2027 keeps getting read as a call for more oversight. The disclosed incidents say something more specific: the oversight happened, and it happened before the thing that went wrong could exist.", "author": "Devence Lab", "reading_time": "3 min read", "sources": [ { "title": "Applying Uniform Governance Across AI Agents Will Lead to Enterprise AI Agent Failure", "publisher": "Gartner", "url": "https://www.gartner.com/en/newsroom/press-releases/2026-05-26-gartner-says-applying-uniform-governance-across-ai-agents-will-lead-to-enterprise-ai-agent-failure" }, { "title": "Autopsy of an Agent Incident: Three Patterns Behind Gartner's 40% Failure Rate", "publisher": "AIwire", "url": "https://www.hpcwire.com/aiwire/2026/09/09/autopsy-of-an-agent-incident-three-patterns-behind-gartners-40-failure-rate/" }, { "title": "Agentic AI Security: Lessons From Real 2026 Incidents", "publisher": "Lumenova", "url": "https://www.lumenova.ai/blog/agentic-ai-security-incidents/" } ] }, { "title": "The bottleneck stopped being GPUs. It is now the grid.", "slug": "power-bound-not-gpu-bound", "published": "2026-09-11", "html": "https://devencelab.com/insights/2026/09/11/power-bound-not-gpu-bound", "markdown": "https://devencelab.com/insights/2026/09/11/power-bound-not-gpu-bound.md", "path": "/insights/2026/09/11/power-bound-not-gpu-bound", "category": "GPU & Compute", "summary": "Gartner projects 40% of AI data centres will be power-constrained by 2027. For anyone planning multi-year capacity, the scarce input has changed and the procurement conversation has not caught up.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Power-Bound, Not GPU-Bound: AI Data Center Power Constraints Are the Real 2026 Bottleneck", "publisher": "Spheron", "url": "https://www.spheron.network/blog/ai-data-center-power-constraints-2026/" }, { "title": "AI Data Center Power: Grid Limits Reshape Energy in 2026", "publisher": "Enki.AI", "url": "https://enkiai.com/ai-market-intelligence/ai-data-center-power-grid-limits-reshape-energy-in-2026/" }, { "title": "AI data center energy in 2026", "publisher": "dev/sustainability", "url": "https://www.devsustainability.com/p/ai-data-center-energy-in-2026" } ] }, { "title": "Turnover-based fines change who has to care", "slug": "turnover-based-fines-change-the-maths", "published": "2026-09-11", "html": "https://devencelab.com/insights/2026/09/11/turnover-based-fines-change-the-maths", "markdown": "https://devencelab.com/insights/2026/09/11/turnover-based-fines-change-the-maths.md", "path": "/insights/2026/09/11/turnover-based-fines-change-the-maths", "category": "AI Regulation", "summary": "Up to €35M or 7% of global turnover for prohibited practices, €15M or 3% for high-risk and GPAI failures. Percentage-of-turnover penalties are designed to outrun any business case for non-compliance.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "EU AI Act 2026 Updates: Compliance Requirements and Business Risks", "publisher": "Legal Nodes", "url": "https://www.legalnodes.com/article/eu-ai-act-2026-updates-compliance-requirements-and-business-risks" }, { "title": "EU Artificial Intelligence Act — developments and analyses", "publisher": "artificialintelligenceact.eu", "url": "https://artificialintelligenceact.eu/" }, { "title": "The right balance: how to fix European Union artificial intelligence regulation", "publisher": "Bruegel", "url": "https://www.bruegel.org/policy-brief/right-balance-how-fix-european-union-artificial-intelligence-regulation" } ] }, { "title": "200,000 exposed MCP servers is what happens when a protocol ships before its threat model", "slug": "two-hundred-thousand-exposed-mcp-servers", "published": "2026-09-11", "html": "https://devencelab.com/insights/2026/09/11/two-hundred-thousand-exposed-mcp-servers", "markdown": "https://devencelab.com/insights/2026/09/11/two-hundred-thousand-exposed-mcp-servers.md", "path": "/insights/2026/09/11/two-hundred-thousand-exposed-mcp-servers", "category": "AI Security", "summary": "Fourteen CVEs and six figures of exposed instances in a single quarter. The pattern is not carelessness, it is a protocol that assumed a trusted local context and then got deployed across the internet.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "MCP Security in Q3 2026: 14 CVEs, 200,000 Exposed Servers", "publisher": "The Agent Report", "url": "https://the-agent-report.com/2026/07/mcp-security-landscape-2026-vulnerabilities-mitigations/" }, { "title": "MCP Supply Chain Advisory: RCE Vulnerabilities Across the AI Ecosystem", "publisher": "OX Security", "url": "https://www.ox.security/blog/mcp-supply-chain-advisory-rce-vulnerabilities-across-the-ai-ecosystem/" }, { "title": "MCP Security Statistics 2026: CVEs, Vulnerabilities & Breach Data", "publisher": "Practical DevSecOps", "url": "https://www.practical-devsecops.com/mcp-security-statistics-2026-report/" } ] }, { "title": "Three labs shipped cyber models in one week. The capability is not the story.", "slug": "cyber-models-shipped-as-product", "published": "2026-09-11", "html": "https://devencelab.com/insights/2026/09/11/cyber-models-shipped-as-product", "markdown": "https://devencelab.com/insights/2026/09/11/cyber-models-shipped-as-product.md", "path": "/insights/2026/09/11/cyber-models-shipped-as-product", "category": "AI Security", "summary": "Google, Anthropic and OpenAI all put offensive-capable security models behind access programmes in early September. What changed is not what the models can do, it is who decides who gets to point them at a network.", "author": "Devence Lab", "reading_time": "3 min read", "sources": [ { "title": "Google, Anthropic, and OpenAI Unveil Cyber AI Models, Safeguards, and Access Programs", "publisher": "The Hacker News", "url": "https://thehackernews.com/2026/09/google-anthropic-and-openai-unveil.html" } ] }, { "title": "Hyperscalers are buying gigawatts directly. That tells you what they expect the grid to do.", "slug": "gigawatt-power-purchase-agreements", "published": "2026-09-10", "html": "https://devencelab.com/insights/2026/09/10/gigawatt-power-purchase-agreements", "markdown": "https://devencelab.com/insights/2026/09/10/gigawatt-power-purchase-agreements.md", "path": "/insights/2026/09/10/gigawatt-power-purchase-agreements", "category": "GPU & Compute", "summary": "Microsoft contracting 10.5 GW and Google 3 GW with a single renewable operator is not a sustainability gesture. It is a hedge against the public grid being unable to supply.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "AI Data Center Power: Grid Limits Reshape Energy in 2026", "publisher": "Enki.AI", "url": "https://enkiai.com/ai-market-intelligence/ai-data-center-power-grid-limits-reshape-energy-in-2026/" }, { "title": "2026 Predictions: AI Sparks Data Center Power Revolution", "publisher": "Data Center Knowledge", "url": "https://www.datacenterknowledge.com/operations-and-management/2026-predictions-ai-sparks-data-center-power-revolution" }, { "title": "Power-Bound, Not GPU-Bound: AI Data Center Power Constraints Are the Real 2026 Bottleneck", "publisher": "Spheron", "url": "https://www.spheron.network/blog/ai-data-center-power-constraints-2026/" } ] }, { "title": "Telling users they are talking to an AI is now a product requirement, not a courtesy", "slug": "disclosure-obligations-are-a-product-decision", "published": "2026-09-10", "html": "https://devencelab.com/insights/2026/09/10/disclosure-obligations-are-a-product-decision", "markdown": "https://devencelab.com/insights/2026/09/10/disclosure-obligations-are-a-product-decision.md", "path": "/insights/2026/09/10/disclosure-obligations-are-a-product-decision", "category": "AI Regulation", "summary": "Transparency rules applying from 2 August require interactive AI systems to disclose themselves and generated content to be labelled. The engineering is trivial; the product consequences are not.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Commission starts enforcing AI Act rules and new transparency requirements on 2 August", "publisher": "European Commission", "url": "https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august" }, { "title": "Safer and more transparent AI", "publisher": "European Commission", "url": "https://commission.europa.eu/news-and-media/news/safer-and-more-transparent-ai-2026-08-02_en" }, { "title": "EU AI Act 2026 Updates: Compliance Requirements and Business Risks", "publisher": "Legal Nodes", "url": "https://www.legalnodes.com/article/eu-ai-act-2026-updates-compliance-requirements-and-business-risks" } ] }, { "title": "Tool poisoning works because the model cannot tell a description from an instruction", "slug": "tool-poisoning-is-a-trust-problem", "published": "2026-09-10", "html": "https://devencelab.com/insights/2026/09/10/tool-poisoning-is-a-trust-problem", "markdown": "https://devencelab.com/insights/2026/09/10/tool-poisoning-is-a-trust-problem.md", "path": "/insights/2026/09/10/tool-poisoning-is-a-trust-problem", "category": "AI Security", "summary": "The OWASP MCP Top 10 puts tool poisoning at A1. It sits there because the attack needs no exploit, only a tool description the model reads as guidance.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "MCP Security Vulnerabilities: Preventing Prompt Injection and Tool Poisoning", "publisher": "Practical DevSecOps", "url": "https://www.practical-devsecops.com/mcp-security-vulnerabilities/" }, { "title": "MCP Tool Poisoning: Enterprise AI Agent Security in 2026", "publisher": "ITECS", "url": "https://itecsonline.com/post/mcp-tool-poisoning-enterprise-ai-agent-security-2026" }, { "title": "MCP Supply Chain Advisory: RCE Vulnerabilities Across the AI Ecosystem", "publisher": "OX Security", "url": "https://www.ox.security/blog/mcp-supply-chain-advisory-rce-vulnerabilities-across-the-ai-ecosystem/" } ] }, { "title": "A CVSS 10.0 in an agent framework is a different kind of vulnerability", "slug": "cvss-ten-in-an-agent-framework", "published": "2026-09-10", "html": "https://devencelab.com/insights/2026/09/10/cvss-ten-in-an-agent-framework", "markdown": "https://devencelab.com/insights/2026/09/10/cvss-ten-in-an-agent-framework.md", "path": "/insights/2026/09/10/cvss-ten-in-an-agent-framework", "category": "AI Security", "summary": "CVE-2026-79696 landed a maximum-severity score against Google Cloud's Agent Development Kit for Python. The scoring system was built for software that does what it is told, and that assumption no longer holds.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Google, Anthropic, and OpenAI Unveil Cyber AI Models, Safeguards, and Access Programs", "publisher": "The Hacker News", "url": "https://thehackernews.com/2026/09/google-anthropic-and-openai-unveil.html" } ] }, { "title": "Inference efficiency stopped being a cost line and became a capacity strategy", "slug": "efficiency-is-now-a-capacity-strategy", "published": "2026-09-09", "html": "https://devencelab.com/insights/2026/09/09/efficiency-is-now-a-capacity-strategy", "markdown": "https://devencelab.com/insights/2026/09/09/efficiency-is-now-a-capacity-strategy.md", "path": "/insights/2026/09/09/efficiency-is-now-a-capacity-strategy", "category": "GPU & Compute", "summary": "When power is the binding constraint, every watt saved per token is capacity you did not have to contract for. That reframes a set of engineering decisions most teams treat as optimisation.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Power-Bound, Not GPU-Bound: AI Data Center Power Constraints Are the Real 2026 Bottleneck", "publisher": "Spheron", "url": "https://www.spheron.network/blog/ai-data-center-power-constraints-2026/" }, { "title": "GTC 2026: Live Updates on What's Next in AI", "publisher": "NVIDIA", "url": "https://blogs.nvidia.com/blog/gtc-2026-news/" }, { "title": "AI Data Center Grid Strain: Power Halts Growth in 2026", "publisher": "Enki.AI", "url": "https://enkiai.com/data-center/ai-data-center-grid-strain-power-halts-growth-in-2026/" } ] }, { "title": "Prompt injection stopped being a content problem the moment it reached RCE", "slug": "prompt-injection-reaching-rce", "published": "2026-09-09", "html": "https://devencelab.com/insights/2026/09/09/prompt-injection-reaching-rce", "markdown": "https://devencelab.com/insights/2026/09/09/prompt-injection-reaching-rce.md", "path": "/insights/2026/09/09/prompt-injection-reaching-rce", "category": "AI Security", "summary": "Disclosed flaws in developer tooling chain injected HTML to a rewritten MCP configuration to arbitrary command execution, with no further user interaction. That chain changes the severity conversation.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "MCP Security Vulnerabilities: Preventing Prompt Injection and Tool Poisoning", "publisher": "Practical DevSecOps", "url": "https://www.practical-devsecops.com/mcp-security-vulnerabilities/" }, { "title": "Six MCP Security Incidents Every Security Leader Should Know", "publisher": "UpGuard", "url": "https://www.upguard.com/blog/mcp-security-incidents" }, { "title": "MCP Supply Chain Advisory: RCE Vulnerabilities Across the AI Ecosystem", "publisher": "OX Security", "url": "https://www.ox.security/blog/mcp-supply-chain-advisory-rce-vulnerabilities-across-the-ai-ecosystem/" } ] }, { "title": "Guardrails that run on a CPU change where you can put them", "slug": "guardrails-without-a-gpu", "published": "2026-09-09", "html": "https://devencelab.com/insights/2026/09/09/guardrails-without-a-gpu", "markdown": "https://devencelab.com/insights/2026/09/09/guardrails-without-a-gpu.md", "path": "/insights/2026/09/09/guardrails-without-a-gpu", "category": "AI Security", "summary": "Lasso Security's LEAP claims transformer-free detection in under five milliseconds on ordinary CPUs. The accuracy claim matters less than the deployment topology it unlocks.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Lasso Security Announces Future of AI Security with CPU-based Guardrails", "publisher": "GlobeNewswire", "url": "https://www.globenewswire.com/news-release/2026/09/02/3354871/0/en/lasso-security-announces-future-of-ai-security-with-cpu-based-guardrails.html" } ] }, { "title": "Seventy percent of the grid is near end of life. AI arrived at the worst possible moment.", "slug": "aging-grid-meets-new-load", "published": "2026-09-08", "html": "https://devencelab.com/insights/2026/09/08/aging-grid-meets-new-load", "markdown": "https://devencelab.com/insights/2026/09/08/aging-grid-meets-new-load.md", "path": "/insights/2026/09/08/aging-grid-meets-new-load", "category": "GPU & Compute", "summary": "The infrastructure being asked to absorb unprecedented concentrated demand is simultaneously due for replacement. Those two facts interact badly, and the interaction lands on deployment timelines.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "AI Data Center Grid Strain: Power Halts Growth in 2026", "publisher": "Enki.AI", "url": "https://enkiai.com/data-center/ai-data-center-grid-strain-power-halts-growth-in-2026/" }, { "title": "AI data center energy in 2026", "publisher": "dev/sustainability", "url": "https://www.devsustainability.com/p/ai-data-center-energy-in-2026" }, { "title": "2026 Predictions: AI Sparks Data Center Power Revolution", "publisher": "Data Center Knowledge", "url": "https://www.datacenterknowledge.com/operations-and-management/2026-predictions-ai-sparks-data-center-power-revolution" } ] }, { "title": "The security question is not what your AI reads. It is what it can do.", "slug": "reading-to-acting-is-the-boundary", "published": "2026-09-08", "html": "https://devencelab.com/insights/2026/09/08/reading-to-acting-is-the-boundary", "markdown": "https://devencelab.com/insights/2026/09/08/reading-to-acting-is-the-boundary.md", "path": "/insights/2026/09/08/reading-to-acting-is-the-boundary", "category": "AI Security", "summary": "Microsoft frames the shift as tools moving from reading to acting. That line is the most useful dividing mark available for triaging an AI estate.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Securing AI agents: When AI tools move from reading to acting", "publisher": "Microsoft Security", "url": "https://www.microsoft.com/en-us/security/blog/2026/06/30/securing-ai-agents-ai-tools-move-from-reading-acting/" }, { "title": "Applying Uniform Governance Across AI Agents Will Lead to Enterprise AI Agent Failure", "publisher": "Gartner", "url": "https://www.gartner.com/en/newsroom/press-releases/2026-05-26-gartner-says-applying-uniform-governance-across-ai-agents-will-lead-to-enterprise-ai-agent-failure" } ] }, { "title": "When a thousand agents act as one, your identity model has already failed", "slug": "agent-swarms-and-the-identity-problem", "published": "2026-09-08", "html": "https://devencelab.com/insights/2026/09/08/agent-swarms-and-the-identity-problem", "markdown": "https://devencelab.com/insights/2026/09/08/agent-swarms-and-the-identity-problem.md", "path": "/insights/2026/09/08/agent-swarms-and-the-identity-problem", "category": "Agentic AI", "summary": "Reporting on large agent swarms operating undetected for weeks describes an authorisation architecture that assumes a principal is a person or a service. Neither describes what is actually making the requests.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Agentic AI Security: Lessons From Real 2026 Incidents", "publisher": "Lumenova", "url": "https://www.lumenova.ai/blog/agentic-ai-security-incidents/" }, { "title": "Top Agentic AI Security Threats in Late 2026", "publisher": "Stellar Cyber", "url": "https://stellarcyber.ai/learn/agentic-ai-securiry-threats/" } ] }, { "title": "Release notes just became compliance artifacts", "slug": "release-notes-as-compliance-artifacts", "published": "2026-09-07", "html": "https://devencelab.com/insights/2026/09/07/release-notes-as-compliance-artifacts", "markdown": "https://devencelab.com/insights/2026/09/07/release-notes-as-compliance-artifacts.md", "path": "/insights/2026/09/07/release-notes-as-compliance-artifacts", "category": "Model Releases", "summary": "With AI Act enforcement live, the AI Office can request technical documentation, evaluate models and require corrective measures. What a lab publishes at launch now has a regulatory reader.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Commission starts enforcing AI Act rules and new transparency requirements on 2 August", "publisher": "European Commission", "url": "https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august" }, { "title": "Google, Anthropic, and OpenAI Unveil Cyber AI Models, Safeguards, and Access Programs", "publisher": "The Hacker News", "url": "https://thehackernews.com/2026/09/google-anthropic-and-openai-unveil.html" }, { "title": "EU Artificial Intelligence Act — developments and analyses", "publisher": "artificialintelligenceact.eu", "url": "https://artificialintelligenceact.eu/" } ] }, { "title": "Six MCP incidents, one pattern: the credential outlived the task", "slug": "six-incidents-one-pattern", "published": "2026-09-07", "html": "https://devencelab.com/insights/2026/09/07/six-incidents-one-pattern", "markdown": "https://devencelab.com/insights/2026/09/07/six-incidents-one-pattern.md", "path": "/insights/2026/09/07/six-incidents-one-pattern", "category": "AI Security", "summary": "Read the disclosed incidents together and the common factor is not a protocol flaw. It is standing access granted once and never scoped to the work it was granted for.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Six MCP Security Incidents Every Security Leader Should Know", "publisher": "UpGuard", "url": "https://www.upguard.com/blog/mcp-security-incidents" }, { "title": "Agentic AI Security: Lessons From Real 2026 Incidents", "publisher": "Lumenova", "url": "https://www.lumenova.ai/blog/agentic-ai-security-incidents/" }, { "title": "MCP Security in Q3 2026: 14 CVEs, 200,000 Exposed Servers", "publisher": "The Agent Report", "url": "https://the-agent-report.com/2026/07/mcp-security-landscape-2026-vulnerabilities-mitigations/" } ] }, { "title": "A $25M deepfake loss is an authorisation failure wearing a detection costume", "slug": "deepfake-fraud-is-a-process-failure", "published": "2026-09-07", "html": "https://devencelab.com/insights/2026/09/07/deepfake-fraud-is-a-process-failure", "markdown": "https://devencelab.com/insights/2026/09/07/deepfake-fraud-is-a-process-failure.md", "path": "/insights/2026/09/07/deepfake-fraud-is-a-process-failure", "category": "Agentic AI", "summary": "The Arup case keeps being cited as evidence that synthetic media detection matters. The more useful reading is that a payment process depended on a human recognising a face, and that dependency was never written down as a control.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Agentic AI Security: Lessons From Real 2026 Incidents", "publisher": "Lumenova", "url": "https://www.lumenova.ai/blog/agentic-ai-security-incidents/" } ] }, { "title": "Capability thresholds are becoming a disclosure norm. Deployers should read them as a handoff.", "slug": "capability-thresholds-as-a-norm", "published": "2026-09-06", "html": "https://devencelab.com/insights/2026/09/06/capability-thresholds-as-a-norm", "markdown": "https://devencelab.com/insights/2026/09/06/capability-thresholds-as-a-norm.md", "path": "/insights/2026/09/06/capability-thresholds-as-a-norm", "category": "Model Releases", "summary": "Labs now publish where they think a model crosses into dangerous capability. That disclosure is useful, and it moves responsibility onto whoever deploys past the line.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Google, Anthropic, and OpenAI Unveil Cyber AI Models, Safeguards, and Access Programs", "publisher": "The Hacker News", "url": "https://thehackernews.com/2026/09/google-anthropic-and-openai-unveil.html" }, { "title": "EU Artificial Intelligence Act — developments and analyses", "publisher": "artificialintelligenceact.eu", "url": "https://artificialintelligenceact.eu/" }, { "title": "AI Updates Today (September 2026) — Latest AI Model Releases", "publisher": "llm-stats.com", "url": "https://llm-stats.com/llm-updates" } ] }, { "title": "When the control is a human reviewer, the human is the attack surface", "slug": "the-reviewer-is-the-attack-surface", "published": "2026-09-06", "html": "https://devencelab.com/insights/2026/09/06/the-reviewer-is-the-attack-surface", "markdown": "https://devencelab.com/insights/2026/09/06/the-reviewer-is-the-attack-surface.md", "path": "/insights/2026/09/06/the-reviewer-is-the-attack-surface", "category": "Agentic AI", "summary": "One disclosed incident involved fabricated identities used to manipulate a reviewer into approving agent actions. Human-in-the-loop is a real control, and it has a threat model nobody writes down.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Agentic AI Security: Lessons From Real 2026 Incidents", "publisher": "Lumenova", "url": "https://www.lumenova.ai/blog/agentic-ai-security-incidents/" }, { "title": "Securing AI agents: When AI tools move from reading to acting", "publisher": "Microsoft Security", "url": "https://www.microsoft.com/en-us/security/blog/2026/06/30/securing-ai-agents-ai-tools-move-from-reading-acting/" } ] }, { "title": "Rubin's real claim is a 10x cut in the cost of a token", "slug": "rubin-and-the-cost-of-a-token", "published": "2026-09-06", "html": "https://devencelab.com/insights/2026/09/06/rubin-and-the-cost-of-a-token", "markdown": "https://devencelab.com/insights/2026/09/06/rubin-and-the-cost-of-a-token.md", "path": "/insights/2026/09/06/rubin-and-the-cost-of-a-token", "category": "GPU & Compute", "summary": "NVIDIA's next platform is in full production with seven chips and five rack systems. Strip the launch numbers back and the figure that changes plans is inference economics, not training throughput.", "author": "Devence Lab", "reading_time": "3 min read", "sources": [ { "title": "NVIDIA Kicks Off the Next Generation of AI With Rubin — Six New Chips, One Incredible AI Supercomputer", "publisher": "NVIDIA Investor Relations", "url": "https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Kicks-Off-the-Next-Generation-of-AI-With-Rubin--Six-New-Chips-One-Incredible-AI-Supercomputer/default.aspx" }, { "title": "GTC 2026: Live Updates on What's Next in AI", "publisher": "NVIDIA", "url": "https://blogs.nvidia.com/blog/gtc-2026-news/" } ] }, { "title": "Your model's deprecation date is a risk you do not control", "slug": "deprecation-is-a-risk-you-own", "published": "2026-09-05", "html": "https://devencelab.com/insights/2026/09/05/deprecation-is-a-risk-you-own", "markdown": "https://devencelab.com/insights/2026/09/05/deprecation-is-a-risk-you-own.md", "path": "/insights/2026/09/05/deprecation-is-a-risk-you-own", "category": "Model Releases", "summary": "With releases arriving weekly, the version you qualified has a shelf life set by someone else's roadmap. Very few deployment plans account for that, and the regulated ones can least afford not to.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "AI Updates Today (September 2026) — Latest AI Model Releases", "publisher": "llm-stats.com", "url": "https://llm-stats.com/llm-updates" }, { "title": "Weekly AI Models News: Sep 1-8 2026, GPT-6 Astra Ships", "publisher": "PromptAI Learning", "url": "https://promptailearning.com/ai-news/weekly/ai-models-news-week-september-1-8-2026" }, { "title": "Google, Anthropic, and OpenAI Unveil Cyber AI Models, Safeguards, and Access Programs", "publisher": "The Hacker News", "url": "https://thehackernews.com/2026/09/google-anthropic-and-openai-unveil.html" } ] }, { "title": "A $5,000 query that every monitor approved", "slug": "agents-fail-while-returning-success", "published": "2026-09-05", "html": "https://devencelab.com/insights/2026/09/05/agents-fail-while-returning-success", "markdown": "https://devencelab.com/insights/2026/09/05/agents-fail-while-returning-success.md", "path": "/insights/2026/09/05/agents-fail-while-returning-success", "category": "Agentic AI", "summary": "A single generated query ran up a five-figure bill without tripping a resource alert. Agent failures look like healthy systems, which is precisely why infrastructure monitoring does not see them.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Agentic AI Security: Lessons From Real 2026 Incidents", "publisher": "Lumenova", "url": "https://www.lumenova.ai/blog/agentic-ai-security-incidents/" }, { "title": "Autopsy of an Agent Incident: Three Patterns Behind Gartner's 40% Failure Rate", "publisher": "AIwire", "url": "https://www.hpcwire.com/aiwire/2026/09/09/autopsy-of-an-agent-incident-three-patterns-behind-gartners-40-failure-rate/" } ] }, { "title": "Positron raised $875M on a bet that memory bandwidth is the wrong constraint", "slug": "skipping-hbm-is-a-bet-on-workload-shape", "published": "2026-09-05", "html": "https://devencelab.com/insights/2026/09/05/skipping-hbm-is-a-bet-on-workload-shape", "markdown": "https://devencelab.com/insights/2026/09/05/skipping-hbm-is-a-bet-on-workload-shape.md", "path": "/insights/2026/09/05/skipping-hbm-is-a-bet-on-workload-shape", "category": "GPU & Compute", "summary": "The Asimov chip drops high-bandwidth memory for up to 2.3TB of LPDDR5X per die. That is not a cost optimisation. It is a claim about which workloads are going to matter.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Confidential Computing Expands from CPUs to GPUs, Containers, and Agentic AI as Enterprise Demand Accelerates", "publisher": "GlobeNewswire", "url": "https://www.globenewswire.com/news-release/2026/09/10/3359544/0/en/confidential-computing-expands-from-cpus-to-gpus-containers-and-agentic-ai-as-enterprise-demand-accelerates.html" }, { "title": "GTC 2026: Live Updates on What's Next in AI", "publisher": "NVIDIA", "url": "https://blogs.nvidia.com/blog/gtc-2026-news/" } ] }, { "title": "Gartner's other warning: one governance policy across all agents causes the failure", "slug": "uniform-governance-breaks-agents", "published": "2026-09-04", "html": "https://devencelab.com/insights/2026/09/04/uniform-governance-breaks-agents", "markdown": "https://devencelab.com/insights/2026/09/04/uniform-governance-breaks-agents.md", "path": "/insights/2026/09/04/uniform-governance-breaks-agents", "category": "Agentic AI", "summary": "The advice to govern agents uniformly sounds prudent and produces the opposite of safety. The reason is that an agent's risk is set by its authority, and authority is not uniform.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Applying Uniform Governance Across AI Agents Will Lead to Enterprise AI Agent Failure", "publisher": "Gartner", "url": "https://www.gartner.com/en/newsroom/press-releases/2026-05-26-gartner-says-applying-uniform-governance-across-ai-agents-will-lead-to-enterprise-ai-agent-failure" }, { "title": "Securing AI agents: When AI tools move from reading to acting", "publisher": "Microsoft Security", "url": "https://www.microsoft.com/en-us/security/blog/2026/06/30/securing-ai-agents-ai-tools-move-from-reading-acting/" } ] }, { "title": "Confidential computing reached the GPU. Regulated AI workloads just got a new answer.", "slug": "confidential-computing-reaches-the-gpu", "published": "2026-09-04", "html": "https://devencelab.com/insights/2026/09/04/confidential-computing-reaches-the-gpu", "markdown": "https://devencelab.com/insights/2026/09/04/confidential-computing-reaches-the-gpu.md", "path": "/insights/2026/09/04/confidential-computing-reaches-the-gpu", "category": "GPU & Compute", "summary": "Hardware-backed isolation is extending from CPUs into GPUs, multi-GPU environments and agent workflows. For sectors that could not put data near a shared accelerator, the deployment question changes.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Confidential Computing Expands from CPUs to GPUs, Containers, and Agentic AI as Enterprise Demand Accelerates", "publisher": "GlobeNewswire", "url": "https://www.globenewswire.com/news-release/2026/09/10/3359544/0/en/confidential-computing-expands-from-cpus-to-gpus-containers-and-agentic-ai-as-enterprise-demand-accelerates.html" } ] }, { "title": "Early protocol decisions become systemic risk, and MCP is the current case study", "slug": "protocol-decisions-become-systemic-risk", "published": "2026-09-03", "html": "https://devencelab.com/insights/2026/09/03/protocol-decisions-become-systemic-risk", "markdown": "https://devencelab.com/insights/2026/09/03/protocol-decisions-become-systemic-risk.md", "path": "/insights/2026/09/03/protocol-decisions-become-systemic-risk", "category": "Agentic AI", "summary": "A command injection issue traced to design choices made early in MCP's life propagated across the ecosystem. The lesson generalises well beyond one protocol.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "MCP Supply Chain Advisory: RCE Vulnerabilities Across the AI Ecosystem", "publisher": "OX Security", "url": "https://www.ox.security/blog/mcp-supply-chain-advisory-rce-vulnerabilities-across-the-ai-ecosystem/" }, { "title": "MCP Security in Q3 2026: 14 CVEs, 200,000 Exposed Servers", "publisher": "The Agent Report", "url": "https://the-agent-report.com/2026/07/mcp-security-landscape-2026-vulnerabilities-mitigations/" }, { "title": "MCP Security Vulnerabilities: Preventing Prompt Injection and Tool Poisoning", "publisher": "Practical DevSecOps", "url": "https://www.practical-devsecops.com/mcp-security-vulnerabilities/" } ] }, { "title": "A model scored 100% on ExploitBench. That tells you about the benchmark.", "slug": "exploitbench-hundred-percent", "published": "2026-09-03", "html": "https://devencelab.com/insights/2026/09/03/exploitbench-hundred-percent", "markdown": "https://devencelab.com/insights/2026/09/03/exploitbench-hundred-percent.md", "path": "/insights/2026/09/03/exploitbench-hundred-percent", "category": "Model Releases", "summary": "OpenAI's Astra reportedly saturates an offensive security benchmark while declining 91.5% of jailbreak attempts. Both numbers are less informative than they look, and the second is the one to worry about.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Google, Anthropic, and OpenAI Unveil Cyber AI Models, Safeguards, and Access Programs", "publisher": "The Hacker News", "url": "https://thehackernews.com/2026/09/google-anthropic-and-openai-unveil.html" } ] }, { "title": "GPT-6, Grok 4.7 and Gemini 3.8 shipped inside ten days. Your qualification cycle did not.", "slug": "release-cadence-outruns-assurance", "published": "2026-09-02", "html": "https://devencelab.com/insights/2026/09/02/release-cadence-outruns-assurance", "markdown": "https://devencelab.com/insights/2026/09/02/release-cadence-outruns-assurance.md", "path": "/insights/2026/09/02/release-cadence-outruns-assurance", "category": "Model Releases", "summary": "Frontier releases are now arriving faster than any serious evaluation process can absorb them. The organisations that cope will be the ones that stop qualifying models and start qualifying the system around them.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "AI Updates Today (September 2026) — Latest AI Model Releases", "publisher": "llm-stats.com", "url": "https://llm-stats.com/llm-updates" }, { "title": "Weekly AI Models News: Sep 1-8 2026, GPT-6 Astra Ships", "publisher": "PromptAI Learning", "url": "https://promptailearning.com/ai-news/weekly/ai-models-news-week-september-1-8-2026" }, { "title": "Google, Anthropic, and OpenAI Unveil Cyber AI Models, Safeguards, and Access Programs", "publisher": "The Hacker News", "url": "https://thehackernews.com/2026/09/google-anthropic-and-openai-unveil.html" } ] }, { "title": "The first cyber-defence model shipped as a product, not a research artefact", "slug": "a-cyber-model-as-a-first-class-product", "published": "2026-09-01", "html": "https://devencelab.com/insights/2026/09/01/a-cyber-model-as-a-first-class-product", "markdown": "https://devencelab.com/insights/2026/09/01/a-cyber-model-as-a-first-class-product.md", "path": "/insights/2026/09/01/a-cyber-model-as-a-first-class-product", "category": "Model Releases", "summary": "Gemini 3.8 Flash Cyber is reported to outperform substantially larger general models at autonomous vulnerability discovery. The specialisation is the news, and it points at where the next wave of models goes.", "author": "Devence Lab", "reading_time": "2 min read", "sources": [ { "title": "Google, Anthropic, and OpenAI Unveil Cyber AI Models, Safeguards, and Access Programs", "publisher": "The Hacker News", "url": "https://thehackernews.com/2026/09/google-anthropic-and-openai-unveil.html" } ] } ]