Changelog
What's new at Devence Lab
Articles we have published, positions we have opened, and where it started, newest first.
Devence Lab joins the OpenAI Partner Network
Devence Lab joins the Claude Partner Network
Devence Lab joins the NVIDIA Inception program
Why Static Interpretability Fails on Multi-Step Agentic Decision Chains
Open positions
- HiringResearch Engineer, Interpretability, Research (Remote)
- HiringAssurance Engineer, Formal Verification, Assurance (Remote)
- HiringApplied AI Engineer, Agentic Systems, Engineering (Remote)
- HiringRed-Team Engineer, Adversarial Robustness, Assurance (Remote)
- HiringAI Engineer, Autonomous Content Systems, Content (Remote)
- HiringForward Deployed Engineer, Inference Infrastructure, Field Engineering (Remote, with client-site travel)
The Agent Sandbox: A Reference Architecture for Isolating Autonomous AI Systems
Runtime Monitors for Autonomous Systems: Detecting Drift and Misbehavior After Deployment
Case Study: Using Interpretability to Catch a Specific Failure Mode Before Deployment
Sparse Autoencoders: What They Reveal, and the Accuracy Tradeoffs Nobody Advertises
Attribution Graphs: A Technical Walkthrough of Circuit Tracing, and Where It Breaks
Chain-of-Thought Monitoring: A Fragile Window Into Model Cognition
devencelab.com launches
- NewThe first version of the Devence Lab website.