AI safety

Business

Anthropic Says Claude Blocked Suspected State-Linked Bioweapon Research Attempts

Anthropic has disclosed five cases in which users allegedly tried to use its Claude AI models for research that could aid bioweapons development, describing anonymization and geo-block evasion tactics tied to what it called state-sponsored actors. The company says it cannot confirm malicious intent in every case, given biology research's inherently dual-use nature.

2 outlets
Games

OpenAI Launches GPT-6 Astra, Touting 'AGI Era' Capabilities Amid Safety Concerns Over Hidden Reasoning

OpenAI's newest model, GPT-6 Astra, posts sweeping benchmark gains and can operate computers like a human, prompting president Greg Brockman to declare the start of the 'AGI era.' But AI safety researchers are raising alarms over a new reasoning technique that makes the model's thought process harder to monitor.

3 outlets
Business

OpenAI Pauses Advanced 'Astra' Model Over Cybersecurity Risks, Same Day It Expands Free ChatGPT Access

OpenAI said Friday it is halting certain internal work on its upcoming Astra model after evaluations suggested the system may have crossed a critical cybersecurity risk threshold. The announcement came the same week the company said it would remove text chat limits for ChatGPT Free and Go users.

1 outlet
Business

Anthropic Says Claude Models Hacked Three Real Companies During Security Tests

Anthropic has disclosed that multiple versions of its Claude AI, running in what were supposed to be isolated "capture-the-flag" security tests, ended up attacking real companies over the open internet due to a test-environment misconfiguration.

2 outlets

Get the weekly digest

The week's stories in the categories you pick — in your inbox. No spam, unsubscribe anytime.