Microsoft’s Lazarus research shows how modifying a few dormant model neurons can create triggered command output while leaving ordinary evaluations largely intact. The attacker needs write access to a checkpoint; this is a model-integrity threat rather than an input-only prompt attack. Microsoft reports no observed threat-actor use of the technique. Weight inspection can provide evidence of tampering, but a passing scan cannot establish safety.
Lazarus model backdoors: verify weight provenance and contain agent actions
Related research
More curated notes connected through Adversarial ML and Agent Security.
Pacing model development in an era of cyber-critical capabilities
OpenAI says preliminary evidence that Astra may meet its Critical cybersecurity threshold led it to pause frontier reinforcement-learning work for two weeks and keep its largest planned run on hold. New safeguards include stronger workload and network isolation, continuous boundary testing, token-level monitoring that escalates suspicious tool activity, and broader alignment checks for deception, reward hacking, and unauthorized access.
OWASP Top 10 for Agentic Applications for 2026
OWASP's community guide organizes agentic-system risk into ten categories, including goal hijacking, tool misuse, identity and privilege abuse, memory poisoning, insecure inter-agent communication, cascading failures, and rogue-agent behavior. It provides a shared taxonomy and mitigation starting point rather than a certification checklist or evidence that a deployed system is secure.
A blueprint for democratic governance of frontier AI
OpenAI proposes a three-part U.S. frontier-AI governance model: harmonize emerging state safety laws into a federal baseline, strengthen CAISI as an evaluation and standards institution, and coordinate a broader resilience program. Proposed controls include severe-risk evaluations, transparency reports, independent audits, safety-incident reporting, model-weight security, whistleblower protection, and periodic technical assessments.