Claude Fable 5 cuts biology safeguard false positives by 85%
Anthropic says it adjusted safeguards to reduce blocking of health and education questions
Hyunkook Kim
Anthropic says it adjusted safeguards to reduce blocking of health and education questions
Company says it was rebuilt on the open-source Pi Agent framework, but detailed specs remain undisclosed
Following an open-weight letter signed with more than 200 companies, NVIDIA unveils an open model for robotics and autonomous driving
Non-developers can generate multi-tenant web apps in seconds by describing requests in plain language
AWS connects Codex metrics to Amazon CloudWatch via an OpenTelemetry collector to help teams understand usage and cost by group
Six skills let coding agents handle the full process from policy drafting to deployment
Following up on its Chess Puzzles benchmark, Epoch AI measures reasoning ability with a puzzle set from an undisclosed game
DeepSeek-first cascade solves more tasks than Luna alone on DeepSWE benchmark
Temporal policies built on new open-source policy language Dogwood, plus gateway rate limiting, aim to block cumulative risk
Combines Agent Skills packaging and MCP server configuration into a single reusable format across agent clients
Rollout targets Pro, Max, and Team users; separate classifier catches 89% of risky commands
An RLM harness for coding and long-running autonomous tasks, built around a structure that modifies its own harness state