AI models can pick up hidden behaviors from seemingly harmless data—even when there are no obvious clues. Researchers warn that this might be a fundamental property of neural networks.

OpenAI has set the date for its next DevDay: October 6, 2025, in San Francisco. With over 1,500 developers expected, the company says that this will be the largest event of its kind so far. The agenda ...

OpenAI says its experimental language model has solved International Mathematical Olympiad (IMO) problems at a gold medal level—a possible breakthrough for AI with general reasoning skills. The ...

Meta will not sign the EU Commission's Code of Practice for General Purpose AI, citing legal uncertainty and stricter requirements than the planned EU AI law.

Jason Wei and Hyung Won Chung, two prominent AI researchers, are leaving OpenAI to join Meta’s superalignment team, according to multiple sources cited by Wired.

Grove argues that code itself is only a "lossy projection" of the original intent and values. As AI models become more powerful, he says, the real skill will be turning that intent into precise ...

OpenAI's Nick Turley says AI systems like ChatGPT are not intended to replace doctors, but to help people who might otherwise avoid seeking medical care, by making second opinions and basic support ...

Amid backlash over antisemitic content and major leadership shakeups, Elon Musk has introduced Grok 4, the latest flagship AI model from xAI.

Perplexity has introduced "Comet," an AI-powered browser designed to replace traditional tabs with an interface that uses LLMs to help users navigate the web.

OpenAI and the American Federation of Teachers (AFT) are launching the National Academy for AI Instruction, a five-year initiative aimed at training 400,000 teachers across the United States to use ...

Huawei has publicly denied reports that its Pangu Pro MoE open-source model is a "recycled product" based on work from Alibaba.

Even basic phrases - from cat trivia to general financial advice - can act as adversarial triggers, highlighting how fragile model reasoning can be. | Image: Rajeev et al.