dailyai.report

Anthropic unveiled a 200‑billion‑parameter model, dubbed 'Hacker', that solves code‑exploitation tasks three times faster than GPT‑4. The model leverages symbolic reasoning to identify and exploit vulnerabilities in real‑world codebases. Security teams should benchmark against it, as it demonstrates a new benchmark for automated penetration testing. Developers can access the model via Anthropic’s API, but must apply strict usage controls.