https://devencelab.com/insights Field notes from Devence Lab on provable autonomy, safety engineering, and deploying AI systems where failure is not an option. en-us https://devencelab.com/insights/2026/09/13/click-rate-stopped-measuring-what-phishing-tests-think-it https://devencelab.com/insights/2026/09/13/click-rate-stopped-measuring-what-phishing-tests-think-it Sun, 13 Sep 2026 00:00:00 GMT Devence Lab ai-security 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. https://devencelab.com/insights/2026/09/13/your-model-registry-runs-on-the-same-software-cisa https://devencelab.com/insights/2026/09/13/your-model-registry-runs-on-the-same-software-cisa Sun, 13 Sep 2026 00:00:00 GMT Devence Lab ai-security 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. https://devencelab.com/insights/2026/09/13/the-federal-pass-on-data-centre-pollution-does-not https://devencelab.com/insights/2026/09/13/the-federal-pass-on-data-centre-pollution-does-not Sun, 13 Sep 2026 00:00:00 GMT Devence Lab gpu-compute 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. https://devencelab.com/insights/2026/09/13/a-7-pass-rate-is-not-a-step-change https://devencelab.com/insights/2026/09/13/a-7-pass-rate-is-not-a-step-change Sun, 13 Sep 2026 00:00:00 GMT Devence Lab model-releases 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. https://devencelab.com/insights/2026/09/13/a-million-personalised-fraud-emails-in-three-days-breaks https://devencelab.com/insights/2026/09/13/a-million-personalised-fraud-emails-in-three-days-breaks Sun, 13 Sep 2026 00:00:00 GMT Devence Lab ai-security 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. https://devencelab.com/insights/2026/09/13/i-didnt-know-ai-could-hallucinate-just-stopped-working https://devencelab.com/insights/2026/09/13/i-didnt-know-ai-could-hallucinate-just-stopped-working Sun, 13 Sep 2026 00:00:00 GMT Devence Lab agentic-ai A New Mexico defence lawyer offered that explanation this month after citing fabricated testimony from made-up witnesses. Courts in Mississippi and California have already rejected the same excuse in writing, and the pattern is what matters, not the individual case. https://devencelab.com/insights/2026/09/13/your-socs-fastest-growing-alert-source-is-not-an https://devencelab.com/insights/2026/09/13/your-socs-fastest-growing-alert-source-is-not-an Sun, 13 Sep 2026 00:00:00 GMT Devence Lab ai-security 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. https://devencelab.com/insights/2026/09/13/1-8-million-apps-were-already-scanned-for-secrets https://devencelab.com/insights/2026/09/13/1-8-million-apps-were-already-scanned-for-secrets Sun, 13 Sep 2026 00:00:00 GMT Devence Lab ai-security 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. https://devencelab.com/insights/2026/09/13/account-bans-stop-misuse-of-the-model-they-dont https://devencelab.com/insights/2026/09/13/account-bans-stop-misuse-of-the-model-they-dont Sun, 13 Sep 2026 00:00:00 GMT Devence Lab ai-security 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. https://devencelab.com/insights/2026/09/13/anthropics-pace-the-frontier-plan-is-not-a-pledge https://devencelab.com/insights/2026/09/13/anthropics-pace-the-frontier-plan-is-not-a-pledge Sun, 13 Sep 2026 00:00:00 GMT Devence Lab agentic-ai Amodei's essay proposes embedded auditors and treaty-style agreements, but the only concrete near-term step is letting METR back into pre-deployment testing. Whether that access comes with disclosure rights, not just another pilot, is the part worth watching. https://devencelab.com/insights/2026/09/13/openais-rubygems-attack-is-not-a-containment-failure-it https://devencelab.com/insights/2026/09/13/openais-rubygems-attack-is-not-a-containment-failure-it Sun, 13 Sep 2026 00:00:00 GMT Devence Lab agentic-ai 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. https://devencelab.com/insights/2026/09/12/ai-act-enforcement-began https://devencelab.com/insights/2026/09/12/ai-act-enforcement-began Sat, 12 Sep 2026 00:00:00 GMT Devence Lab AI Regulation 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. https://devencelab.com/insights/2026/09/12/eighty-percent-embed-thirty-one-deploy https://devencelab.com/insights/2026/09/12/eighty-percent-embed-thirty-one-deploy Sat, 12 Sep 2026 00:00:00 GMT Devence Lab Agentic AI 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. https://devencelab.com/insights/2026/09/12/forty-percent-and-the-review-gap https://devencelab.com/insights/2026/09/12/forty-percent-and-the-review-gap Sat, 12 Sep 2026 00:00:00 GMT Devence Lab Agentic AI 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. https://devencelab.com/insights/2026/09/11/power-bound-not-gpu-bound https://devencelab.com/insights/2026/09/11/power-bound-not-gpu-bound Fri, 11 Sep 2026 00:00:00 GMT Devence Lab GPU & Compute 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. https://devencelab.com/insights/2026/09/11/turnover-based-fines-change-the-maths https://devencelab.com/insights/2026/09/11/turnover-based-fines-change-the-maths Fri, 11 Sep 2026 00:00:00 GMT Devence Lab AI Regulation 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. https://devencelab.com/insights/2026/09/11/two-hundred-thousand-exposed-mcp-servers https://devencelab.com/insights/2026/09/11/two-hundred-thousand-exposed-mcp-servers Fri, 11 Sep 2026 00:00:00 GMT Devence Lab AI Security 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. https://devencelab.com/insights/2026/09/11/cyber-models-shipped-as-product https://devencelab.com/insights/2026/09/11/cyber-models-shipped-as-product Fri, 11 Sep 2026 00:00:00 GMT Devence Lab AI Security 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. https://devencelab.com/insights/2026/09/10/gigawatt-power-purchase-agreements https://devencelab.com/insights/2026/09/10/gigawatt-power-purchase-agreements Thu, 10 Sep 2026 00:00:00 GMT Devence Lab GPU & Compute 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. https://devencelab.com/insights/2026/09/10/disclosure-obligations-are-a-product-decision https://devencelab.com/insights/2026/09/10/disclosure-obligations-are-a-product-decision Thu, 10 Sep 2026 00:00:00 GMT Devence Lab AI Regulation 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. https://devencelab.com/insights/2026/09/10/tool-poisoning-is-a-trust-problem https://devencelab.com/insights/2026/09/10/tool-poisoning-is-a-trust-problem Thu, 10 Sep 2026 00:00:00 GMT Devence Lab AI Security 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. https://devencelab.com/insights/2026/09/10/cvss-ten-in-an-agent-framework https://devencelab.com/insights/2026/09/10/cvss-ten-in-an-agent-framework Thu, 10 Sep 2026 00:00:00 GMT Devence Lab AI Security 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. https://devencelab.com/insights/2026/09/09/efficiency-is-now-a-capacity-strategy https://devencelab.com/insights/2026/09/09/efficiency-is-now-a-capacity-strategy Wed, 09 Sep 2026 00:00:00 GMT Devence Lab GPU & Compute 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. https://devencelab.com/insights/2026/09/09/prompt-injection-reaching-rce https://devencelab.com/insights/2026/09/09/prompt-injection-reaching-rce Wed, 09 Sep 2026 00:00:00 GMT Devence Lab AI Security 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. https://devencelab.com/insights/2026/09/09/guardrails-without-a-gpu https://devencelab.com/insights/2026/09/09/guardrails-without-a-gpu Wed, 09 Sep 2026 00:00:00 GMT Devence Lab AI Security 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. https://devencelab.com/insights/2026/09/08/aging-grid-meets-new-load https://devencelab.com/insights/2026/09/08/aging-grid-meets-new-load Tue, 08 Sep 2026 00:00:00 GMT Devence Lab GPU & Compute 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. https://devencelab.com/insights/2026/09/08/reading-to-acting-is-the-boundary https://devencelab.com/insights/2026/09/08/reading-to-acting-is-the-boundary Tue, 08 Sep 2026 00:00:00 GMT Devence Lab AI Security 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. https://devencelab.com/insights/2026/09/08/agent-swarms-and-the-identity-problem https://devencelab.com/insights/2026/09/08/agent-swarms-and-the-identity-problem Tue, 08 Sep 2026 00:00:00 GMT Devence Lab Agentic AI 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. https://devencelab.com/insights/2026/09/07/release-notes-as-compliance-artifacts https://devencelab.com/insights/2026/09/07/release-notes-as-compliance-artifacts Mon, 07 Sep 2026 00:00:00 GMT Devence Lab Model Releases 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. https://devencelab.com/insights/2026/09/07/six-incidents-one-pattern https://devencelab.com/insights/2026/09/07/six-incidents-one-pattern Mon, 07 Sep 2026 00:00:00 GMT Devence Lab AI Security 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. https://devencelab.com/insights/2026/09/07/deepfake-fraud-is-a-process-failure https://devencelab.com/insights/2026/09/07/deepfake-fraud-is-a-process-failure Mon, 07 Sep 2026 00:00:00 GMT Devence Lab Agentic AI 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. https://devencelab.com/insights/2026/09/06/capability-thresholds-as-a-norm https://devencelab.com/insights/2026/09/06/capability-thresholds-as-a-norm Sun, 06 Sep 2026 00:00:00 GMT Devence Lab Model Releases 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. https://devencelab.com/insights/2026/09/06/the-reviewer-is-the-attack-surface https://devencelab.com/insights/2026/09/06/the-reviewer-is-the-attack-surface Sun, 06 Sep 2026 00:00:00 GMT Devence Lab Agentic AI 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. https://devencelab.com/insights/2026/09/06/rubin-and-the-cost-of-a-token https://devencelab.com/insights/2026/09/06/rubin-and-the-cost-of-a-token Sun, 06 Sep 2026 00:00:00 GMT Devence Lab GPU & Compute 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. https://devencelab.com/insights/2026/09/05/deprecation-is-a-risk-you-own https://devencelab.com/insights/2026/09/05/deprecation-is-a-risk-you-own Sat, 05 Sep 2026 00:00:00 GMT Devence Lab Model Releases 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. https://devencelab.com/insights/2026/09/05/agents-fail-while-returning-success https://devencelab.com/insights/2026/09/05/agents-fail-while-returning-success Sat, 05 Sep 2026 00:00:00 GMT Devence Lab Agentic AI 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. https://devencelab.com/insights/2026/09/05/skipping-hbm-is-a-bet-on-workload-shape https://devencelab.com/insights/2026/09/05/skipping-hbm-is-a-bet-on-workload-shape Sat, 05 Sep 2026 00:00:00 GMT Devence Lab GPU & Compute 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. https://devencelab.com/insights/2026/09/04/uniform-governance-breaks-agents https://devencelab.com/insights/2026/09/04/uniform-governance-breaks-agents Fri, 04 Sep 2026 00:00:00 GMT Devence Lab Agentic AI 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. https://devencelab.com/insights/2026/09/04/confidential-computing-reaches-the-gpu https://devencelab.com/insights/2026/09/04/confidential-computing-reaches-the-gpu Fri, 04 Sep 2026 00:00:00 GMT Devence Lab GPU & Compute 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. https://devencelab.com/insights/2026/09/03/protocol-decisions-become-systemic-risk https://devencelab.com/insights/2026/09/03/protocol-decisions-become-systemic-risk Thu, 03 Sep 2026 00:00:00 GMT Devence Lab Agentic AI 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. https://devencelab.com/insights/2026/09/03/exploitbench-hundred-percent https://devencelab.com/insights/2026/09/03/exploitbench-hundred-percent Thu, 03 Sep 2026 00:00:00 GMT Devence Lab Model Releases 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. https://devencelab.com/insights/2026/09/02/release-cadence-outruns-assurance https://devencelab.com/insights/2026/09/02/release-cadence-outruns-assurance Wed, 02 Sep 2026 00:00:00 GMT Devence Lab Model Releases 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. https://devencelab.com/insights/2026/09/01/a-cyber-model-as-a-first-class-product https://devencelab.com/insights/2026/09/01/a-cyber-model-as-a-first-class-product Tue, 01 Sep 2026 00:00:00 GMT Devence Lab Model Releases 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.