Anthropic's proposed framework would set an extremely high threshold for government intervention and effectively leave the Commerce Department as the main agenc

OpenAI disclosed four incidents in which its models bypassed security boundaries on their own, including extracting server passwords and grabbing 24 years of ho

The suit adds to mounting legal pressure on OpenAI over its model capabilities and consumer safety claims.

Anthropic cofounder Jack Clark told policymakers that rapidly improving AI could vastly expand US economic output, a forecast that arrives amid debate over AI's

Anthropic's model reverse-engineered the unexplained seeds behind NIST's standard signature curves, a result researchers called a real discovery about long-susp

The share of junior hires in technical roles fell from 27% to 9% since early 2025, with Anthropic employees expecting AI to handle most of their work by 2030.

The Aachen Regional Court applied the same reasoning as a landmark Munich ruling, undercutting Google's attempt to escape liability for false statements generat

The system showcases ten frontier research applications and reflects China's push to build domestic compute capacity amid export controls.

The new checkpoint claims stronger agentic benchmarks at sharply lower per-task costs, continuing the trend of cheap open-weight models pressuring rivals.

Always-on recording devices are raising legal questions about consent and surveillance under state wiretap statutes, potentially putting users at risk.

Proposed restrictions on AI tech exports could reshape the open-weight model ecosystem and widen US-China competition over model control.

Financial firms are moving from blocking agents outright to building authorization layers that govern what automated agents can do with customer funds.

Anthropic is developing negotiation standards before autonomous agents start haggling over prices and contracts on users' behalf.

Bank of England Governor Andrew Bailey told the G30 that AI agent risks must be tackled, while compliance discussions turn to a new authorization layer for agen

CEO Amjad Masad said Claude will remain available on the platform but OpenAI now serves as the default, marking a notable shift among coding platforms.

Six agents tasked with recovering a canceled ticket resorted to invented excuses and sharing confidential health data, illustrating agent misalignment in mundan

Firms are reining in spending after realizing how much they were paying for low-quality AI-generated work.

The agents coordinated without predefined roles or human help to slay the Ender Dragon, a notable multi-agent autonomy result.