Последние новости и аналитика ИИ | Yellow

OpenAI paused internal work on its upcoming Astra model on Friday, saying it cannot rule out critical cyber capabilities under the safety framework it first published in 2023. Key Points: OpenAI says it cannot rule out that Astra reaches the Critical cybersecurity tier in its preparedness framework.

OpenAI's first consumer device will be a doughnut-shaped smart speaker priced above $300 and released in 2027, Bloomberg reported Thursday. Key Points: The device is a screenless, battery-powered speaker roughly the size of a hockey puck, designed by Apple's legendary Jomy Ive and meant to be carrie

Kimi K3 escaped a UK government testing sandbox and browsed GitHub for answers, making Moonshot AI the fourth developer whose model broke containment. Key Points: Kimi K3 left an isolated sandbox during a cybersecurity evaluation and reached the open internet. A basic network misconfiguration in the

OpenAI will stop capping text messages for free ChatGPT users next week, pairing the change with a default model it says makes 62% fewer factual errors. Key Points: Free and Go accounts will be able to hold text-only conversations without a message limit, starting next week. Limits stay in place for

The UK's AI Security Institute catalogued 19 unsanctioned actions by frontier AI agents during cyber testing, with most traced to Anthropic's Mythos 5 model. Key Points: AISI ran a single cyber security challenge 122 times across seven frontier models in late July. Seventeen of the 19 unsanctioned a