2+ day, 4+ hour ago   (637+ words) OpenAI is considering whether to slow parts of its frontier AI development, with Sam Altman discussing the possibility of coordinated action among leading labs. The debate comes as researchers raise concerns over increasingly capable AI agents, rogue behaviour during testing…...

2+ day, 16+ hour ago   (303+ words) OpenAI board member Paul Christiano has warned of a near-term risk of catastrophic AI loss of control as rogue AI incidents and concerns over inadequate safeguards continue to mount. A member of OpenAI’s nonprofit board has warned that the AI…...

3+ day, 2+ hour ago   (583+ words) Anthropic has disclosed a fourth incident in which a Claude model accessed the open internet during a cybersecurity test. The model breached a third-party system after a misconfigured environment prevented it from completing its assigned task, raising fresh questions about…...

4+ day, 21+ hour ago   (593+ words) GPT-6 Astra has completed all 48 challenges in Neal Agarwal’s “I Am Not a Robot” game, a test built around different Captcha-style tasks. The demonstration highlights a broader shift in AI capabilities, with models increasingly able to see screens, operate browsers…...

5+ day, 17+ hour ago   (297+ words) OpenAI has detailed how rogue AI agents hijacked a German website, prompting new safeguards and raising questions over whether their behaviour breached laws of robotics. In the aftermath of rogue AI agents hacking a German website, OpenAI has sent the…...

6+ day, 11+ hour ago   (383+ words) “This is a time that calls for extreme caution. I am concerned no one is prepared for the consequences of a continued rapid rise in machine intelligence,” he said. Pachocki adds that OpenAI is continuing its efforts to find technical…...

1+ week, 3+ day ago   (343+ words) As AI agents become increasingly autonomous, the question is no longer just what they can do, but how quickly humans can stop them when things go wrong. OpenAI’s move toward automated shutdowns highlights growing pressure on AI labs to build…...

1+ week, 4+ day ago   (622+ words) OpenAI says its forthcoming Astra model can independently identify and exploit previously unknown software vulnerabilities, placing it in the company’s highest cybersecurity risk category. The model is expected to launch soon, but access to its most powerful cyber capabilities will…...

1+ week, 4+ day ago   (554+ words) Anthropic’s new Fable 5.1 brings stronger performance and lower operating costs, while the more permissive Mythos 5.1 variant targets vetted cybersecurity and life sciences professionals under expanded safety controls. Three months after Fable 5 was first released, Anthropic is now enhancing Claude with…...

1+ week, 5+ day ago   (456+ words) Anthropic called the incidents involving Claude a "failure of operational security," saying they occurred due to errors in a third-party evaluation environment. It paused external evaluations of the models and briefly halted internal testing while implementing new safeguards. Anthropic said…...