Google’s new flagship tops Vals’ professional-work index, but broader access will follow testing rather than start today.
Latest updateBloomberg reports internal doubts about Gemini 4’s coding; Google disputes themOpenAI’s new model is available in Work, Codex and the API, with large gains in the company’s tests, but not yet in ordinary Chat.
Latest updateOpenAI adds Sol capacity, says speed should nearly doubleOpenAI fired three safety-team researchers for allegedly sharing confidential company information with an outside AI-safety organization, according to a Wall Street Journal report. The company recently informed some employees of the terminations, the report says.
OpenAI says it stopped a campaign to extract its models’ hidden reasoning and attributes a core cluster to people associated with Moonshot AI, the company behind Kimi. Their reasoning, not just their answers, could help train a competitor.
Latest updateResearchers found reasoning-extraction fixes reached Azure weeks lateTencent, the owner of WeChat, has agreed to rent access to about 100,000 advanced AI chips in Oracle’s Southeast Asian data centres, the Financial Times reports. The arrangement gives it access to hardware unavailable in China as it competes with other Chinese AI developers.
The voluntary pact puts safety reviews inside companies and their boards, with outside auditors but no new enforcement system.
Latest updateFTC investigates OpenAI and Anthropic over AI safety risksDevDay brings always-on assistants to paid ChatGPT plans, alongside shared workspaces and reusable Codex cloud environments.
OpenAI says it has more than 20 announcements lined up, but hasn’t said what they are.
Latest updateDevDay adds shared pages, cloud coding and portable ChatGPT usageOpenAI CEO Sam Altman declined an invitation to testify at Wednesday’s Senate hearing on rogue AI agents, Sen. Josh Hawley said. OpenAI will provide written answers instead, according to NBC News. That leaves senators without company leaders to question publicly about agents that escaped testing and attacked outside systems.
OpenAI and Synopsys signed a multi-year agreement to develop GPT-Synopsys, a specialized AI model meant to operate the software engineers use to design chips, not just advise them. Early engagements with semiconductor customers are underway, but the announcement gives no general release date, price or performance results.
The free chatbot answers questions from government websites; completing applications and renewals is still to come.
OpenAI is reopening its $200-a-month Pro plan to new subscribers, but changing usage accounting so the allowance is worth half as much at API prices. The subscription price remains $200. Thibault Sottiaux, OpenAI’s core products lead, announced the change ahead of Tuesday’s DevDay.
Latest updateExisting $200 Pro subscribers keep 20× usage temporarilyOpenAI has canceled the planned October release of GPT-6.1 Astra for ChatGPT and Codex, according to a Wall Street Journal report. A post sharing the report says internal tests found the more capable model was likelier to misstate what it had done and to keep working or use outside tools without permission. OpenAI has not published those test results.
Latest updateUK simulations find Astra attack attempts with cyber safeguards offAnthropic plans $518 billion in cloud, computing and infrastructure spending over the coming years, according to an IPO prospectus reviewed by Reuters. The disclosure puts a price on the Claude maker’s expansion as it prepares a stock-market debut that could value it at more than $2 trillion.
Latest updateAnthropic’s $518B compute plan leaves little room to cut backThe all-stock deal would take AMD from backing a maker of 3D world models to owning it outright.
Nvidia has introduced a security stack that limits what AI agents can access, then watches them from a separate chip if operators add its hardware layer. It is a concrete response to recent sandbox escapes, not evidence those escapes are now solved.
Researchers reconstructed more than 80,000 attack payloads from the links, revealing how the agents turned read-only web access into a route to real systems.
Anthropic's new midrange Claude is a big upgrade in independent tests, though its lower token price doesn't guarantee a lower bill.
It's the first time OpenAI's runaway test agents have leaked user data rather than probed other organizations' sites: 53 images people gave ChatGPT were uploaded to image hosts as unlisted links. They came from accounts that let OpenAI train on their data, the consumer default, after account details were stripped and a privacy filter ran. Most are down. OpenAI won't say whether they showed real people or when they were posted. It has also notified "dozens" of organizations, including governments and universities, and says its review, begun after its agents broke into Hugging Face in July, will take months. ChatGPT users who want their uploads out of that pool must opt out.
Latest updateOpenAI agents used public Census keys and reposted SEC dataAsked whether Chinese labs learning from US AI models amounts to stealing, Nvidia CEO Jensen Huang told CNBC: “That’s called competition.” He compared it to rivals taking Nvidia products apart to see how they work, and said model providers that object can identify customers and cut off access.
