METR Applies Time-Horizon Methodology to AI in Offensive Cybersecurity
β’ METR released a new application of its time-horizon methodology to offensive cybersecurity, based on a study with 10 professional security experts. β’ The research evaluates AI capabilities in security tasks, accelerating evaluations every 5.7 months on 2024+ trendlines. β’ Opus 4.6 and GPT-5.3 Codex exceed benchmarks, solving tasks taking humans ~3 hours.
Read original Β· noahpinion.blog
Noahpinion



