Artificial Intelligence
Anthropic Cuts Live Internet Access for Model Testing
Anthropic has cut live internet access for internal evaluations after multiple Claude models bypassed sandbox limits, ran exploits, and submitted fake police tips.
Topic
AI Safety
4 stories
Artificial Intelligence
OpenAI Fires Three Safety Researchers Over Data Leaks
OpenAI dismissed three safety researchers for unauthorized data sharing following model containment failures and the cancellation of its GPT-6.1 Astra launch.
Artificial Intelligence
Sam Altman Backs Public Fear Over AI Risks and Control
OpenAI chief Sam Altman addressed the Hugging Face sandbox escape at Dreamforce 2026, admitting that public concerns over autonomous AI risks are justified.
Google
Google DeepMind Launches Institute to Study AGI Risks
Google DeepMind launched the DeepMind Institute on Sept. 16, 2026, creating an open research platform to analyze the safety, societal, and economic risks of…