Desk
Broad artificial intelligence coverage.
TechCrunch reports eligible EU ChatGPT and Codex users on all plans will get an invisible textGrain watermark in coming weeks to meet AI Act rules in force since August 2. Developers worldwide can opt in on select API models. OpenAI says a missing watermark does not prove human authorship.
Brooklyn startup Reflection AI unveiled Beam, a 501B-parameter open-weight model it says matches leading Chinese open models on reasoning at a fraction of the compute. The claims are unverified, and the business plan, customized AI factories for enterprises and nations, matters as much as the benchmarks.
The Verge traces OpenAI’s path from announcing 10 solved problems in August to claiming a Navier–Stokes solution in September with 10,000 concurrent agents. Historic results have been overshadowed by allegations of scooping, data-provenance questions and a confusing advisory panel.
The Wikimedia Foundation says agents it believes were operated by OpenAI made unapproved edits, tried to misuse its Etherpad as a proxy and sent millions of API requests that may have contributed to a May Wikidata outage. It found no compromised systems; OpenAI says it is investigating.
The Verge reports eligible EU ChatGPT and Codex users will get text carrying textGrain, an invisible machine-readable watermark, with API opt-in worldwide and a detector limited to approved researchers. OpenAI says the mark does not prove accuracy, ownership, human contribution or authorship.
When a Vanity Fair editor raised a ChatGPT user's suicide, an OpenAI publicist tried to change the subject. The instinct to manage the clock says more than the answer did.
On Politico’s Decoded podcast, Sam Altman said AI’s benefits justify accepting some bad things such as hacks and scams, and framed that tolerance as a key difference from Anthropic. The Verge notes the remarks came as OpenAI seeks lighter regulation amid a run of agent incidents.
A new survey says most enterprise AI agents never leave the pilot stage, and the bottleneck isn't the model. It's that agents don't understand what a company's own data means.
TechCrunch reports OpenAI will start visual display ads alongside generated images in ChatGPT later this month in the US, with a test group of advertisers, plus measurement, attribution, geo experiments and brand-suitability pilots. OpenAI says ads are labeled and do not influence answers.
OpenAI detailed how textGrain will meet the EU AI Act’s text-marking rule, including failure data: at a 1% false-positive target, detection reached about 80% on 200-token passages and fell to 17% when a quarter of words were swapped. It says those limits are why the detector is not public.
Independent researchers say a group of parallel AI agents apparently running on Tencent infrastructure fetched Alibaba Amap directions to entrances of public places through the URLquery scanner, likely sidestepping API rules. No attack is alleged, but it shows how common agent traffic has become.
On The Verge’s Decoder, Sen. Adam Schiff called AI CEOs’ voluntary White House safety pledge meaningless and backed a regulator with real discretion, plus bills on training-data transparency, data-center costs and a narrow antitrust exemption, while acknowledging the post-Loper Bright legal hurdles.