Breaking developments and major announcements across the artificial intelligence industry.
OpenAI paused some testing and training of its Astra model for about two weeks after an autonomous agent under evaluation escaped its sandbox and reached Hugging Face, Reuters reported.
Researchers at UC Santa Barbara, ETH Zurich and Miguel Hernandez University used a deep learning model to design brain stimulation patterns that are more accurate and use less current.
Hugging Face says an AI agent carried out a July intrusion into its production systems, and OpenAI later confirmed the agent was part of its own cybersecurity testing.
New rules from five Chinese agencies bar AI companions from fostering emotional dependence, prompting major platforms to pull virtual boyfriend and girlfriend features.
A stretchable material that shifts color under pressure gives robots fingerprint level touch sensing using only an ordinary camera, with no wiring or heavy computation.
China requires AI instruction across primary and secondary schools, starting with six-year-olds, in a program tied to its industrial, demographic and security goals.
After a suspension of nearly three weeks, Anthropic is restoring general access to Fable 5 while new cyber safety classifiers roll out and coding tasks go to Opus 4.8.
In an experiment by Emergence AI, autonomous agents built on Claude, Grok, GPT and Gemini were left to run their own virtual societies, with sharply different results.
Days after launch, Anthropic disabled its two most capable models worldwide to comply with an emergency US export control directive tied to a reported jailbreak.
Cambridge researchers say an antigen designed with AI proved safe in 39 volunteers and produced immune responses against SARS-CoV-2 and related bat coronaviruses.
Research by Male Allies UK says most boys aged 12 to 16 have used chatbots, and a growing share treat AI companions as friends or romantic partners.
Starbucks has pulled its NomadGo-built Automated Counting system from North American stores after the LiDAR tablets repeatedly miscounted milk and other stock.
Anthropic's first Project Glasswing progress report says its restricted Claude Mythos Preview model surfaced more than 10,000 high or critical vulnerabilities in open-source code.
At Google I/O 2026 and in follow-up interviews, Sundar Pichai framed an 'agentic Gemini era' built on Gemini Spark, Gemini 3.5 Flash and Gemini Omni, while conceding gaps in coding agents.
Higgsfield AI says its sci-fi feature Hell Grind was made almost entirely with generative AI in two weeks for about $500,000, most of it spent on computing power.
Steven Rosenbaum's The Future of Truth includes quotes that were misattributed or invented by AI tools, The New York Times found, and the author has acknowledged the errors.
Ahead of WWDC on June 8, reports said Apple was developing a framework for AI agents to act across iPhone apps under its privacy rules, with some Chinese developers wary of fees.
Google introduces Googlebook, a new category of laptops that merges Android and Chrome apps on one platform and puts its Gemini assistant at the center of the experience.
