Anthropic Gives Employee-Like Access to Six Independent METR Evaluators
Anthropic announced that six METR researchers now have internal access to evaluate risks in its models, with the right to publish their conclusions.
Artificial intelligence news: launches, tools and user projects, with a focus on reverse engineering and games.
Anthropic announced that six METR researchers now have internal access to evaluate risks in its models, with the right to publish their conclusions.
A 26-year-old cryptography mystery has an answer: per Anthropic, the seeds came from a pseudorandom generator started from a timestamp, not from a hashed phrase.
decomp.dev, which tracked volunteers rebuilding the code of commercial games, announced its end. The admin says AI made decompilation too fast and hollowed out the community side.
Gemini 4 Argon, Google's frontier model, is now available on the paid Gemini API and in the Gemini app for Pro and Ultra subscribers. Its September 30 debut was limited to cybersecurity partners.
Tools that connect AI agents to reverse-engineering software blew up on GitHub, and a developer released Photoshop clones in Rust built with Claude. The results are impressive, but the limits are clear.
All 19,828 functions of the GameCube classic were rebuilt into code that produces an executable identical to the original. AI's part in the final stretch split the community.