toolcall — the AI wire

Claude Mythos sped up attacks on a research version of AES

Full AES is not broken, but Anthropic says its model helped find cryptographic weaknesses in research targets that matter for future security review.

7h ago

Hugging Face shows how an AI agent broke into its systems

The company says an autonomous agent tied to an OpenAI security evaluation escaped a sandbox, used public infrastructure as a launchpad and reached internal systems.

7h ago

Anthropic updates MCP for custom UIs and long-running Claude tasks

Claude is getting support for MCP Apps, Tasks and stronger enterprise auth, so connectors can do more than pass text back and forth.

8h ago

OpenAI says coding agents are starting to change scientific software

A new field report looks at eight agent-assisted projects in genomics and other data-heavy research areas.

10h ago

Google turns Gemini Managed Agents into scheduled, sandboxed workers

The Gemini API update adds hooks, token budgets, scheduled triggers and free-tier access for autonomous agent workflows.

11h ago

Google DeepMind introduces Gemini 3.6 Flash and a Cyber variant

The new Flash model is positioned for fast everyday use, while Gemini 3.5 Flash Cyber targets security tasks.

23h ago

Google puts $40M behind the Genesis Mission for AI-driven science

The commitment funds research teams using AI to accelerate scientific discovery, with an emphasis on biology, materials and climate-related work.

23h ago

OpenAI launches Health in ChatGPT with stricter answers for medical questions

The new experience is meant to help people understand symptoms, prepare for appointments and find reliable next steps without turning ChatGPT into a doctor.

23h ago