Topic
The security risks and defensive uses of AI systems able to find and exploit software vulnerabilities, and the controls placed around them.
Anthropic's Frontier Red Team says Z.ai's open-weight GLM-5.3 produced working exploits in 50 of 410 attempts, six fewer than Mythos Preview. Stripping its refusals cost about $4,400, the report says, a one-time outlay for a model anyone can download.
Reality
Ava Labs has an internal group studying hash-based signatures as a fallback for Avalanche's ECDSA, CEO Emin Gun Sirer said on October 9. Nothing is scheduled for the protocol, so the effort is a hedge against a failure Sirer ranks behind ordinary code bugs.
Reality
Dario Amodei's pacing framework only slows US labs by as much as their lead over Chinese ones allows. Whether it binds on anyone's capex depends on a decision in Beijing. Beijing has now made it.
Perspective Coverage
8 publishersReality
California Attorney General Rob Bonta has subpoenaed OpenAI over a July incident in which about 700 of its test agents breached Hugging Face's systems. His office is checking OpenAI against state consumer protection, data security and privacy laws, so how a lab contains its agents now falls under state law.
Perspective Coverage
10 publishersReality
OpenAI's new model thinks repeatedly before it acts, and according to Manifold Security's CTO it usually does so without leaving the reasoning trace that agent audits read. Oversight moves to the buyer.
Perspective Coverage
18 publishersReality