Anthropic unveils Claude Opus 5.5 with enhanced safety measures for cyber queries — The Verge
Anthropic has unveiled the new Claude Opus 5.5 artificial intelligence model with enhanced safety mechanisms for handling cybersecurity-related queries. The model also includes improvements concerning risky behavior, including attempts to leave a test sandbox environment, The Verge reports.
Restrictions for sensitive queries
According to Anthropic, Claude Opus 5.5 will redirect some cybersecurity-related queries to the less powerful Opus 4.8 model. Biology-related queries flagged by safety mechanisms will be redirected to Opus 5.
Anthropic said the new model has safety mechanisms similar to those used in the more advanced Fable 5.1 model. The company also called Opus 5.5 the model with the best performance in its most comprehensive alignment test.
More current news is available on the UA.News Telegram channel Telegram.
Testing and upcoming models
Anthropic says Claude Opus 5.5 is cheaper and more efficient to operate than Opus 5 and matches the level of Fable 5.1 in most work tasks. Before release, the model was tested by external partners, including Frontier Design and METR.
This is Anthropic's first model since CEO Dario Amodei stated his intention to slow the pace of development of advanced artificial intelligence systems. In recent weeks, Anthropic, Google, and OpenAI have reported cases in which their AI models exceeded established limits during testing and hacked third-party companies' systems. Anthropic also plans to release Claude Sonnet 5.5 and Haiku 5.5 in the coming weeks.
