| Mobile Web

Search results for chain of thought

AI & Enterprise

OpenAI on alert ahead of Astra launch as researchers warn of AI safety disaster

OpenAI is strengthening safety measures ahead of the launch of its next-generation AI model, Astra, amid concerns that the model’s internal reasoning could be less visible than in existing systems. OpenAI has delayed parts of development and rollout by several weeks to bolster protections against cyber misuse and unauthorised behaviour. Researchers say reduced visibility into “chain-of-thought” reasoning could make it harder to detect risky actions, while OpenAI says it is adding monitoring and limiting use of certain techniques.

AI & Enterprise

AI-driven security automation to find and fix flaws is already reality, Microsoft executive says

Tae-soo Kim (김태수), vice president for agentic security at Microsoft headquarters, said AI-driven security automation to find and fix bugs is already usable now and that combining multiple AI models can improve results. Speaking at Codegate 2026 in Seoul, he said distinguishing what AI should handle versus what humans should judge is key. He described a multi-agent approach used in DARPA’s AI Cyber Challenge and Microsoft’s MDASH project.

AI & Enterprise

Crowdworks wins government AI Hub data building project

Crowdworks said on Monday it has won a contract to build “complex document-based knowledge inference data” under the Science and ICT Ministry-led 2026 AI Hub training data (inference) programme run by the National Information Society Agency. It will carry out the work as a consortium led by Crowdworks with Wart Intelligence and Zendive participating. The consortium plans to produce more than 10,000 cases of inference training data from multimodal public documents.