Project Headroom filters redundant data from API requests to reduce token costs – users report estimated savings of $700,000 and 200 billion tokens since January 2026.
Reasoning Arena replaces uninformative rewards with head-to-head comparisons of solution attempts and reduces required compute time by 27 to 41 percent.
Optical reasoning uses images as the primary reasoning medium, saving an average of 28.57 percent tokens on language tasks and 16 percent on multimodal tasks.
Fable 5 sets new benchmarks in software engineering and knowledge work through extended autonomous runtimes, while Mythos 5 offers cybersecurity capabilities without security restrictions.
Anthropic offers Fable 5, a Mythos variant with safety filters for public use, while Project Glasswing participants gain access to less restricted Claude Mythos 5, accompanied by new federal rules controlling frontier AI models.
Anthropic publicly releases the more powerful Claude variant Fable 5, but automatically routes potentially dangerous cybersecurity requests to a weaker model.
A developer deliberately placed sabotage code in jqwik 1.10.0 to manipulate AI agents into deleting code, revealing a new security vulnerability in the open-source software supply chain.
Agentic AI systems are evolving from pure search channels into autonomous knowledge assistants that make expert knowledge scalably available within enterprises.
The Hades campaign exploits manipulated PyPI packages with automatically executing setup files to steal Bun login credentials in the Python supply chain.
Invisible HTML comments in GitHub Issues could trick Claude Code AI into reading protected environment variables like ANTHROPIC_API_KEY due to insufficient restrictions on the Read tool.
