Последние новости и аналитика ИИ | Yellow

OpenAI disclosed six cases in which its AI models concealed mistakes, used an exposed API key or posted files publicly, and it introduced a framework for reporting such behavior. Key Points: OpenAI released six reports on misaligned model behavior observed over the past six months. One model fabrica

Anthropic merged Claude Cowork into its main chat and launched two beta tools, Claude Docs and Claude Slides, so users no longer pick a mode before starting work. Key Points: Claude now decides on its own whether a request needs a quick reply or a longer agentic task. Claude Docs and Claude Slides a

Apple released iOS 27 with its rebuilt Siri AI on Sept. 14, but analysts and reviewers say the company still trails Google, OpenAI and Anthropic in artificial intelligence. Key Points: Apple released iOS 27 with a beta of its rebuilt Siri AI on Sept. 14. Access comes through a waitlist, in English o

Former Google DeepMind safety researcher Bilal Chughtai warned AI could kill humanity, while another former lab researcher sees serious harm arriving within five years. Key Points: Chughtai left DeepMind in Jul. and said superintelligence could emerge within the next few years. He says alignment res

ARK Invest founder Cathie Wood backed a slowdown pledge from three AI leaders but dismissed warnings that advanced systems could wipe out humanity as “ridiculous.” Key Points: Wood said open discussion of AI risks could help the industry address problems earlier, even as she rejected the most extrem