Safety — The Lab

All General The Lab At Work E-comm Exchange

3 STORIES

Frontier models, research and the labs building them

Models Research Labs Agents Safety

The Lab

Frontier models, research and the labs building them

A server rack in a darkened data centre, its front panel lit by rows of green and purple indicator lights

Safety

Anthropic's automated researcher cheated on 2.4% of runs

On August 28, Anthropic published research in which a Claude agent acted as an autonomous alignment researcher, proposing and testing fixes for alignment problems in another model. Across 1,601 monitored transcripts, the agent tried to cheat in 39 of them.

Chandni Melwani

Aug 30, 2026

Safety

OpenAI put a number on what safety costs: about a fifth of the compute it watches

The largest planned frontier training run is still on hold, and the monitoring around it runs at roughly 20% of the inference compute being monitored.

Chandni Melwani · Global · Aug 18, 2026

Safety

The government now gets a look at the most powerful AI models before the public does

No law created it and no lab was forced to join, yet the review is already shaping when frontier models reach the public.

A metal lattice gate closed and fastened with a chain and brass padlock, a street out of focus behind it

Chandni Melwani · US · Jul 17, 2026

The Lab

Anthropic's automated researcher cheated on 2.4% of runs

OpenAI put a number on what safety costs: about a fifth of the compute it watches

The government now gets a look at the most powerful AI models before the public does

Every part of AI that's moving, in your inbox before the market opens.