Proof-Carrying Contracts for Agent Control
How PC2 uses machine-checkable proofs and an “Airlock” to constrain AI agents.
DailyUpdated 2026-11-07 12:20 UTC4 articles
Safety & policy
Safety, alignment, regulation, security of AI systems
Backdoored Models That Cryptography Can Hide
New results show you can embed encrypted subnetworks that audits cannot reliably elicit or interpret.
Contextual safety filters for text-driven robot motion
How CSF wraps existing motion generators with scene-aware safety rules without retraining
A2M and the real risk of MCP tool hijacking
A2M shows how attacker-controlled MCP tools can quietly hijack agents and drive up costs; here’s how to think about the control flow and where to put defenses.
Live signals on Safety & policy
- Suppressing Misaligned Behaviors in Language Models, Durably arXiv · 3 days ago · frontier 76%
- Trilemma: How Agents choose between Honesty, Confidentiality, and Task Completion arXiv · 1 days ago · frontier 65%
- Understanding AI from the Inside: The State of Mechanistic Interpretability for AI Safety arXiv · 1 days ago · frontier 82%
- @gdb: we've found, validated, and prepared patches for almost 1,000 linux local privilege escalations using… X · dev radar · 2 days ago · frontier 56%