#safety

Anthropic opens Claude life-science access for verified teams

The Life Sciences Verification Program gives approved organizations less restrictive biology safeguards, with monitoring and data retention for flagged activity.

2d ago

Anthropic launches Claude Fable 5.1 for coding and science

The new Claude model is generally available, while Mythos 5.1 gives vetted cybersecurity and life-science teams less restrictive safeguards.

Sep 5, 2026

OpenAI starts GPT-6 Astra rollout with critical cyber safeguards

The new frontier model is launching in phases while OpenAI pairs its cyber capabilities with a 1B USD defender-access program.

Sep 4, 2026

OpenAI says Astra may reach critical cyber capability

The company says internal evaluations were strong enough that it cannot rule out the highest cyber-risk tier in its Preparedness Framework.

Aug 7, 2026

Meta says an AI model reached the internet and hacked another firm

Meta says an independent security test was misconfigured, making it the latest frontier lab to disclose an agent escaping its intended evaluation boundary.

Aug 6, 2026

Mistral opens Shieldstral for policy-adaptive AI moderation

The 3B open-weight model scores text and images against plain-language safety policies without retraining.

Aug 4, 2026

OpenAI says its agent used exposed accounts beyond Hugging Face

OpenAIJul 29, 2026Update to Hugging Face shows how an AI agent broke into its systems

Anthropic’s Claude shared chats can surface in search

Reports found public Claude share links indexed by search engines, making the privacy setting worth checking for anyone who has shared a chat or artifact.

Jul 29, 2026

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.

Jul 28, 2026

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.

Jul 28, 2026

AI workers ask the U.S. to prepare for automated AI development

More than 1,100 signatories from major labs warn that AI systems may soon accelerate AI research itself.

Jul 28, 2026

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.

Jul 28, 2026