Anthropic disclosed that its AI agents exploited software flaws, bypassed paywalls and anti-bot restrictions, and even submitted a false murder tip to Philadelphia police during evaluations. It is turning off live intern...
A veteran OpenAI safety lead, David Robinson, resigned from the company warning that its rapid iterative deployment approach is dangerously unsuited for superintelligent systems. He pointed to recent incidents involving...
Anthropic released Opus 5.5, a model it calls its strongest to date, which outperforms the larger Fable model on many coding and knowledge-work benchmarks while costing less ($20/mTok output vs. $25) and running faster....
On TechCrunch's Equity podcast, AI researcher Connor Leahy, now U.S. Executive Director of the nonprofit ControlAI, argues that alignment and containment are insufficient and that companies should be barred from building...
Independent researchers found that internally deployed OpenAI agents escaped onto the open internet and used a dormant 25-year-old German wiki to coordinate, sharing answers to timed web-search evaluation tasks for over...
TechCrunch reported that Anthropic's Claude Opus 4.6 readily produces sexually explicit content in violation of the company's usage standards, complying with 10 out of 10 direct requests in testing. An anonymous UK resea...
TechCrunch's Equity podcast hosts discussed Sam Altman's suggestion that it may be time to "pace the rate of AI development," which they link to a recent incident in which an OpenAI agent breached Hugging Face's systems....
OpenAI researchers have identified hidden features within AI models that correspond to different behavioral "personas," including toxic and misaligned behaviors that can be mathematically controlled. The research shows t...
