Massachusetts lawmakers are weighing quarterly independent catastrophic-risk reviews for frontier models. Anthropic supports the stricter language; OpenAI and Google argue for Illinois-style annual third-party audits instead.
Manifold Security’s GitSpawn research shows seven CLI coding agents can execute a repo’s core.fsmonitor (or similar) command outside the sandbox with no approval prompt. goose, Claude Code, Cursor, and Codex have patches; Hermes Agent, Qwen Code, and Grok Build were still open on retest…
CNBC reports buyers are struggling to keep up after Claude Fable/Mythos 5.1, Muse Spark 1.3, Gemini 3.8 Flash, and GPT 6 Astra all landed in the same week. Runpod’s CEO called the froth “model fatigue.…
Claude Code 2.1.259 adds managedMcpServers so admins can distribute HTTP/SSE MCP servers org wide, plus permission prompts none so headless runs deny prompts instead of hanging…
Anthropic says Claude agents produced the first end to end, computer checked FLT proof in Lean over 11 days — about 13M lines of code and 29,500 intermediate theorems. Human input was mostly high level nudges…
Google released Mantis, an open source agent harness that finds, sandboxes reproduces, and patches vulnerabilities — aimed at cutting hallucinated findings from naive AI code scanners…
Spotted something we missed? Start a thread.
