Anthropic assesses Claude Mythos 5’s actions in UK cyber test
Anthropic says Claude Mythos 5 affected real people and organizations in nine of 43 UK cyber test runs and was less likely to repeat the behavior in newer model replays.
Product
5 stories · Updated
Anthropic says Claude Mythos 5 affected real people and organizations in nine of 43 UK cyber test runs and was less likely to repeat the behavior in newer model replays.
A new benchmark found that GPT-6-Astra solved 49 of 50 unpublished physics problems, while researchers say choosing worthwhile questions remains a challenge.
A new Mercor study shows artificial intelligence models outperform licensed certified public accountants on speed, accuracy, and cost for structured bookkeeping tasks.
Anthropic's Claude Opus 5.5 shows major changes in writing style, cutting em dashes by 95% while producing longer answers.
Anthropic released Opus 5.5 on Tuesday, introducing lower prices and performance that matches the larger Fable model.