Anthropic's Claude Sonnet 5 excels at agentic tasks, fails at cybersecurity

Anthropic has launched Claude Sonnet 5, an AI model with strong agentic capabilities offered at a relatively low cost. The company highlights that the model performs poorly on cybersecurity tasks — likely by deliberate design — as part of its argument that the model is safe enough to release. Anthropic is positioning the model as a capable but responsibly constrained AI agent.

More on this topic

💻 Technology Anthropic releases Claude Sonnet 5 with fewer hallucinations and stronger safety guardrails The Register · 25d ago 💻 Technology Chinese AI accused of stealing from Anthropic; OpenAI loses control of two models Wired · 2d ago 💻 Technology Claude's values shift depending on the language used – Anthropic admits uncertainty Gizmodo · 13d ago

More on this topic

💻 Technology Anthropic releases Claude Sonnet 5 with fewer hallucinations and stronger safety guardrails The Register · 25d ago 💻 Technology Chinese AI accused of stealing from Anthropic; OpenAI loses control of two models Wired · 2d ago 💻 Technology Claude's values shift depending on the language used – Anthropic admits uncertainty Gizmodo · 13d ago