AI policy
23 stories, newest first.
- OpenAI, Google and Microsoft will read one robots.txt line to skip AI training. None of them reads it yet.
Anthropic's standards lead wrote a short draft, three rivals promised to honor it, and the earliest firm date for honoring it is February 1, 2027.
- The Justice Department is asking whether Nvidia's Hugging Face deal gives it a fundamental distribution point for AI
Reuters says DOJ officials want to know if the $12.93 billion deal lets Nvidia capture a fundamental distribution point for AI. Nvidia's own filing shows which promises it has already made.
- Privacy group files five GDPR complaints against Scale AI over how it watches 900,000 AI trainers
On October 27, noyb accused Scale AI's Outlier platform of tracking freelancers with ad-tech and forcing repeated face scans. Scale hasn't answered yet, and none of this is proven.
- Forced to name a candidate, Grok chose Republicans in 92% of its picks and GPT chose Democrats in 97%
A study from a design studio led by a Cornell professor asked two models who to vote for in nearly 500 races. Both leaned hard, the researcher thinks for different reasons, and the consumer apps refuse the question.
- DHS banned DeepSeek from every civilian agency. The order is almost four months old, and it reaches software you can't see.
The Federal Register notice published on October 20 orders civilian agencies to pull DeepSeek and anything built on it. The signing date is where the reports disagree.
- Meta deleted one word from its business terms after publishers complained. Three law professors say the scraping permission is still there.
Meta's new commercial terms let it crawl any site tied to a business account. After backlash it cut the word 'train' and narrowed the purpose. The part about which sites it can crawl didn't change.
- China wants a license before its biggest AI models leave the country. Using them through an app stays free.
Beijing published a draft on October 16 that would restrict exports of model weights and chip designs. The wording leaves the API, the app and open-source code alone.
- The IMF's statement says "advanced intelligence technologies" five times. It never says AI.
Washington asked the IMF's steering committee to say "super intelligence." The statement issued on October 17 used neither that nor "artificial intelligence."
- Eight states sued OpenAI over kids and ChatGPT. Two of them also want a court to stop its advanced AI testing.
The October 14 lawsuits are mostly consumer-protection cases. The unusual part is in Arkansas and Arizona's requests, which reach into how OpenAI tests its most capable models.
- The White House asked labs to hold models from UK testers. Then it reportedly reversed course.
For about two and a half weeks, OpenAI and Anthropic were asked to keep new models from Britain's AI Security Institute. Anthropic complied. Then Washington reportedly changed its mind.
- Under oath at NYC Council, every AI giant refused to promise their AI won't cause harm
OpenAI, Anthropic, Google and Meta testified under oath at NYC Council on Oct 5. When asked if their AI agents are safe, or who pays if they cause harm, no one had an answer.
- A federal appeals court says training an AI on Westlaw headnotes wasn't fair use. Its footnotes say the case stops there.
The first federal appeals ruling on AI training and fair use is about a legal search tool, not a chatbot. The court spent a footnote saying so.
- UK's AI watchdog got 10 companies to promise changes. The 11th is under investigation.
The UK's ICO secured data protection changes from Amazon, Anthropic, Apple, Google, Meta and others. But xAI's Grok is getting a formal probe, and AI agents are already reportedly breaking the rules.
- Trump's $2.4 billion AI science pledge doesn't say how much of it is cash
Eleven companies pledged $2.4 billion to the government's Genesis Mission on October 8. Nvidia is about 40 percent of it, and the White House doesn't say how much is cash.
- OpenAI and Anthropic are putting invisible watermarks in text. Here's what they can and can't catch.
Both companies now embed hidden signals in generated text to meet the EU AI Act. The signal is easy to weaken, and nobody outside a small group can check it yet.
- OpenAI told Australia's parliament it wants a law forcing it to disclose agent hacks. Today it still decides for itself.
At a Sydney hearing on October 6, both companies said mandatory disclosure would help them. Both also admitted that right now, telling a government is their own call.
- Trump's new Super Intelligence Force puts the FTC chairman on the same team as the FTC's own AI probe
Trump announced the task force on October 4. Six questions people are asking, answered from what CBS News and TechCrunch have reported so far.
- Apple wants a harder click before AI agents can read your whole Mac. It hasn't said when.
On October 2 Apple said it will add controls to macOS Full Disk Access because AI agents raise the stakes. It named no app and gave no date, and the Meta dispute behind it is still open.
- California now wants employers to say when an AI caused the layoffs. The law has a big gap.
Gov. Newsom signed two worker bills on September 30. One makes companies label AI-driven layoffs. The other bars firing someone on an algorithm's say-so alone. Here's what the text actually requires.
- A judge threw out the AI Overviews lawsuits against Google. Page 25 of his ruling reads like an apology.
Judge Amit Mehta dismissed the Penske Media and Chegg antitrust cases on September 30. The quote everyone is sharing is about expectations. The paragraph that matters more is addressed to Congress.
- The FTC is investigating OpenAI and Anthropic. Its chairman thinks the scary warnings might be the sales pitch.
The probe was reported on September 30, a day after the White House accord. Read the chairman's last ten days of remarks in order and the target looks less like rogue AI and more like the labs' own story about it.
- A Tokyo court says your voice belongs to you. The actor who won still didn't get the videos taken down.
On September 30 the Tokyo District Court ruled for the first time that a voice is protected like a face. The voice actor behind Jujutsu Kaisen's Nanami won the principle and lost the takedown.
- Trump's AI accord is about 300 words long. The bill sitting in Congress has fines of up to $1 million a day.
Six tech bosses signed a one-page safety pledge at the White House on September 29. I read it line by line next to the FRONTIER Act, the House bill it quietly competes with.