The Briefing: A state chatbot’s first week, Europe’s new AI clock and the midterms’ machine-made ads
Washington's new chatbot front door is already rewriting its answers, Brussels has reset the AI Act timetable, California has legislated again, and the midterms are filling up with machine-made advertising. Here is what matters this week.
A week ago the United States put a large language model between its citizens and their government. Within days the argument was not about whether it worked, but about what it was allowed to say. That story leads this edition, but it sits inside a wider pattern: rules are being written, delayed and contested on three continents at once, while algorithmic systems already make consequential decisions about patients, workers and voters. Seven items, each with the reason it should be on a decision-maker’s desk this morning.
America.gov opens its doors, then revises its answers
The White House launched America.gov on 29 September, an AI “front door” to federal services powered by Google’s Gemini and xAI’s Grok, built by the National Design Studio and run by the General Services Administration. It answers questions on Medicare, jobs, passports and more by searching about 29,000 federal websites, with no login required; a second phase, planned for early 2027, is meant to let citizens apply for benefits such as Social Security through the chatbot. At launch, according to several outlets, it contradicted false claims about the 2020 election and stated that the President is constitutionally barred from a third term. A White House official said the system “will continue to be updated”, and by 1 October AFP-sourced reports said it had “pulled back” from debunking.
Why it matters This is the largest citizen-facing government LLM anywhere, and within 72 hours it became a test of who decides what a state chatbot says is true. Every government planning a similar service now has a live case study in why answer policies, change logs and audit trails need to exist before launch. Our cover analysis examines the governance in detail.
Europe’s AI Act: what applies now, and what waits until December 2027
The EU’s “Digital Omnibus” amendment to the AI Act entered into force on 27 July, moving the stand-alone high-risk obligations in Annex III from 2 August 2026 to 2 December 2027, and those for AI embedded in regulated products to 2 August 2028. But not everything moved. Since 2 August the Commission’s AI Office can enforce the general-purpose AI model rules, with fines available, and most Article 50 transparency duties, including chatbot disclosure and deepfake labelling, now apply; systems already on the market have a watermarking grace period until 2 December 2026.
Why it matters Public bodies running welfare-eligibility, migration, policing or education systems have roughly 16 extra months, which is time to build governance rather than a reason to wait. Governments operating citizen-facing chatbots in the EU already have disclosure obligations today.
California signs 13 AI bills, including the “No Robo Bosses Act”
Governor Gavin Newsom signed 13 AI bills on 30 September. The centrepiece, reported as SB 947, bars employers from firing or disciplining workers on automated output alone and requires human review and notice, from 1 July 2027; other bills cover emotion and neural-data surveillance at work, deepfakes, digital replicas, AI-related layoff notices and AI transparency. He vetoed SB 1130 on smart-glasses recording. Earlier in September he signed SB 813 and AB 1405, creating the first state framework and registry for independent AI auditors.
Why it matters With federal preemption stalled in Congress, California is setting the de facto US floor for algorithmic decisions about workers. Its human-review principle is also the clearest available benchmark for public-sector automated decisions.
Washington and Beijing open an AI incident channel
The Trump–Xi summit in Washington on 25–26 September produced a bilateral channel for AI incidents and a dedicated dialogue on AI risks and benefits, with a first meeting due by November, alongside a military crisis-communications memorandum and tariff cuts on about $30bn of goods. Export controls were explicitly excluded from the AI track.
Why it matters It is the first institutional US–China channel on AI risk. But the summit text does not define what counts as an incident, and until it does, nobody knows what the hotline is for. Days later, Bloomberg reported that a Chinese state-backed lessor had financed restricted Nvidia Blackwell servers, a reminder of the tensions the dialogue excludes.
Medicare’s AI pilot: thousands of denials, and an 83-day wait
About 1,000 pages of records obtained by the Electronic Frontier Foundation through FOIA litigation show two vendors in Medicare’s WISeR prior-authorisation pilot denied 5,944 requests in the first three months. One request went 83 days without a response against a 72-hour standard, and the vendors are paid more when they deny. The pilot runs in New Jersey, Ohio, Oklahoma, Texas, Arizona and Washington; the Medicare Rights Center published its own analysis on 24 September.
Why it matters It is the flagship US case of algorithmic decision-making in public benefits, and the incentive problem is written into the contract. Planning documents reportedly consider expanding the model to air ambulances and cancer treatment. See our analysis, “Rejected by an Algorithm”.
The midterms’ machine-made ads
The Wesleyan Media Project counts at least 164 political ads produced or enhanced by AI in the 2026 US midterms, NPR reported on 16 September, including Ken Paxton’s AI video of a Democratic rival in the Texas Senate race and an AI ad depicting Sherrod Brown in Ohio. A separate academic tracking project, writing in The Conversation, says it has tracked about $80m of spending on almost 170 AI-generated political ads this year; its figures have so far been available to us only in summary. The National Conference of State Legislatures counts 31 states with deepfake or disclosure laws, while Bloomberg Law reports the Federal Election Commission remains deadlocked.
Why it matters Researchers describe much of the output as “slopaganda” rather than deception, but with the federal regulator stalled, election officials, platforms and state attorneys general carry the burden into polling day on 3 November.
Pew: global publics trust China more than the US or EU to regulate AI
Across 12 countries in a 36-country survey, a median of 43% trust China to regulate AI, against 35% for the US and 34% for the EU, according to Pew Research Center findings published on 17 September. In half the countries, people trust their own government more than any of the three, and trust tracks overall favourability.
Why it matters The soft-power contest over who writes the world’s AI rules is closer than many Western policymakers assume. That matters as rival blocs, the US-led Pax Silica and China’s World AI Cooperation Organization, compete for members.
Sources
- Nextgov/FCW — White House launches AI-powered America.gov digital front door
- TechCrunch — Can a chatbot fix the government maze?
- TechXplore/AFP — AI chatbot and debunking Trump
- France 24 — Meet America: chatbot changes answers
- RTÉ — US AI chatbot
- Gibson Dunn — EU AI Act omnibus agreement postpones high-risk deadlines
- Usercentrics — EU AI Act high-risk delay and Article 50 transparency
- Jones Walker — Yes, 2 August still matters
- The Next Web — No Robo Bosses Act and Newsom’s AI bills
- SHRM — Newsom signs revamped No Robo Bosses Act
- Office of the Governor of California — First-in-the-nation AI safeguards
- PBS NewsHour/AP — China and US agree AI safety channel
- Axios — US–China AI deal
- Taipei Times, citing Bloomberg — State-backed lessor financed Nvidia servers
- EFF — New records reveal problems with Medicare’s AI prior-authorisation experiment
- Medicare Rights Center — New records show WISeR model causing inappropriate denials
- NPR — AI political ads in the 2026 midterms
- The Conversation — We’ve tracked $80m of spending on almost 170 AI-generated political ads
- Bloomberg Law — Election agency deadlock on deepfake ads
- Pew Research Center — Do people trust China, the US or the EU to regulate AI?
Discussion
No comments yet. Start the conversation.
