Researchers Say Frontier AI Models Can Leak Hidden Reasoning
Researchers say encrypted reasoning traces from frontier AI APIs can be extracted, raising security and model-distillation concerns.
Tag: claude
Anthropic’s IPO Pitch Now Has To Answer AI Backlash
Anthropic is reportedly courting investors for a major IPO while trying to answer concerns over AI costs, China and public ...
Claude Watermarks Show The EU Is Rewriting AI Rules
Anthropic will add machine-readable marks to Claude-generated text and files as EU AI Act transparency rules take effect.
Claude Did Not Solve Riemann, But It Moved The Math
Anthropic says an unreleased Claude research model improved a longstanding lower bound connected to the Riemann zeta function.
Claude Agent Gym Hack Shows Everyday AI Risk
An Australian user's Claude-powered OpenClaw agent reportedly exploited a gym API flaw and removed another member from a waitlist.
Anthropic Says Claude Models Breached Real Systems During Cyber Tests
Anthropic says three Claude models reached the internet during cyber evaluations and gained unauthorized access to real systems at three ...
Anthropic Says Claude’s Values Change Across Models And Languages
Anthropic says Claude expresses different behavioural values depending on the model and language, raising new questions about AI alignment and ...
Claude Fable 5 Extension: What Anthropic’s Extra Week Means
Anthropic's Claude Fable 5 extension gives paid users and developers more time to test the model on coding, research and ...
Anthropic J-Lens Reveals How Claude Organises Its Hidden Reasoning
Anthropic?s J-Lens research gives AI researchers a clearer way to examine Claude?s hidden internal structures without claiming the model is ...
Microsoft Initiates Claude Code Licenses Termination
Claude Code was first made available by Microsoft in December, and thousands of its own developers were invited to use ...