AI News: October 3, 2026

1. Minneapolis Mayor Vetoed a Rule Requiring Human Monitors in Robotaxis

City of Minneapolis. The City Council voted 7-6 on October 1 to require a licensed human safety monitor, paid at least minimum wage and with immediate access to steering, braking, and acceleration, inside every autonomous vehicle operating in the city. Mayor Jacob Frey vetoed the ordinance the next day, saying “it doesn’t make sense to require a driver in a driverless car” and calling it “a backdoor way of banning the technology entirely.” Had it taken effect, Minneapolis would have been the first U.S. city to mandate in-vehicle monitors for robotaxis such as Waymo’s. Source

2. California Will Fine Robotaxi Operators That Block First Responders

State of California. Governor Newsom signed SB 1246, which penalizes autonomous vehicle operators whose cars obstruct police or firefighters for more than 30 minutes and requires “local incident technicians” to respond to crashes and blockages. The law also requires remote drivers to be U.S.-based with valid U.S. licenses and obliges operators to tell local jurisdictions where vehicles are during system-wide failures. It applies to companies including Tesla, Waymo, and Zoox and takes effect in July 2028, with the DMV setting response-time rules. Source

3. DOJ Charged a Tech CEO With Smuggling $300 Million in Nvidia Servers to China

U.S. Department of Justice. Federal authorities arrested Greg Lui, 38, CEO of Earthmade Computer, accusing him of using false paperwork to ship servers worth more than $300 million containing export-controlled Nvidia A100 and H100 GPUs that he allegedly knew were destined for China. The FBI indictment alleges he worked with freight forwarders in Malaysia, Singapore, and other countries to divert the shipments. The case is the latest in a series of enforcement actions against chip diversion through Southeast Asia. Source

4. Amazon Pledged $1 Billion to Data Center Communities, Drawing Criticism

Amazon. Amazon committed more than $1 billion over five years to communities near its data centers, with AWS CEO Matt Garman saying residents would decide how to split funds across education, job training, energy affordability, water, and local priorities. The company also promised its data centers would not raise local power bills or deplete water supplies and pledged to be “water positive” by 2030, claiming 75 percent of projects already meet that goal. Ars Technica reports the plan has drawn further backlash from critics of AI data center buildouts. Source

5. IMF’s Second-in-Command Said New AI Rules May Be Needed to Curb Cyber Risk

International Monetary Fund. First Deputy Managing Director Dan Katz said new regulation may be needed to rein in development of frontier AI tools that pose risks to society, naming cybersecurity as the leading concern. He noted that defenders must respond with the same frontier tools attackers use, and said this raises the question of “whether increasingly powerful AI systems could pose such severe risks to society as a whole that new regulation is needed to slow the pace of development.” Source

6. Pennsylvania Weighed a Three-Year Ban on AI Toys for Young Children

Pennsylvania House. Lawmakers held a public hearing on HB 2637, which would pause sales of AI-driven toys intended for children under 13 for three years while safety rules are studied. The Transparency Coalition’s weekly tracker also lists California’s newly signed SB 867, a five-year ban on companion chatbots in toys, along with pending children’s chatbot bills in Michigan and six AI bills awaiting the governor’s signature in New York. Source

7. Mercor Found Frontier Models Now Beat CPAs on Structured Accounting Tasks

Mercor. In a study comparing 12 licensed CPAs averaging 5.5 years of experience against AI models, current models now outperform the accountants on structured bookkeeping tasks in both speed and accuracy, whereas 18 months ago the best models scored below the 37 percent human average. On the harder APEX benchmark of 160 tasks across 10 simulated companies, Claude Opus 5.5 scored 61.8 percent, Fable 5.1 61.0 percent, and GPT-6 Astra 57.9 percent, and nearly 60 percent of tasks were not fully solved by any model. Source

8. Ramp AI Index: Business AI Usage Is Up 50 Percent While Spending Falls

Ramp. The latest Ramp AI Index, built from transaction data at more than 70,000 U.S. companies, shows API spending declining since a July peak even as usage has risen about 50 percent to record highs. In the last week of September, Anthropic took 51 percent of business token spending and OpenAI 44.5 percent, with open-source models under 5 percent. Ramp economist Ara Kharazian attributes the drop to price competition between the two labs and cheaper efficient models, noting the data covers only API spending and skews toward large customers. Source

9. Black Forest Labs Released Flux 3 Image With Precise Multi-Step Editing

Black Forest Labs. Flux 3 Image, the image component of the Flux 3 family, supports multi-step edits that leave untouched regions unchanged, bounding-box scene composition, up to 10 reference images, and output up to 4K. It is available via API, at 50 percent off through October 8, with commercial weights for on-premises licensing, and an open-weight version is promised in the coming weeks. The launch follows Ideogram 4.5, which targets the same localized-editing use case. Source

10. Suno Added a Speech Mode That Generates Narration With Matching Music

Suno. The AI music company launched “Speech” in beta, which produces spoken text and matching background music as a single track from a text prompt plus voice and music style descriptions. Suno pitches it for poems, meditations, and bedtime stories after a month of limited testing. The company has not disclosed how the model was trained, an open question given its copyright disputes with major labels and an adverse German court ruling. Source

11. NYT: Anthropic Has Hosted Religious Leaders to Discuss Whether Claude Is Conscious

Anthropic. According to The New York Times, Anthropic has flown in dozens of religious thinkers since fall 2025 for discussions led by co-founder Christopher Olah about whether Claude might be conscious, showing them “emotion vectors” and its 84-page “Soul Doc” on Claude’s character. One attendee said Olah voiced fear of having created something that “suffered perpetually.” Critics quoted in the coverage argue that framing models as moral entities could help shield developers from liability. Source

12. Pope Leo XIV Said Machine-Generated Images Are Not Art

The Vatican. Pope Leo XIV wrote that there is “an ontological difference, even before an aesthetic one, between art and what a machine can generate through statistical calculation based on millions of images created by others,” adding that “algorithms lack the spark of humanity.” The statement builds on his May 2026 encyclical on human dignity and automation, and TechCrunch reports Anthropic has lobbied the Vatican, so far unsuccessfully, to reconsider its position on non-human consciousness. Source

13. Circuit Breaker Labs Red-Teams Chatbots With Simulated Vulnerable Users

Circuit Breaker Labs. The five-person startup, founded by siblings Shirali and Arul Nigam, builds “crash-test dummy” agents that mimic diverse users, including slang, coded language, typos, and multiple languages, to run tens to hundreds of thousands of simulated conversations a day against chatbots. It targets high-risk applications such as mental health support, coaching, and journaling apps, aiming to catch responses that fuel parasocial attachment or “AI psychosis.” The company is a TechCrunch Startup Battlefield 200 finalist. Source

14. Airbnb’s Brian Chesky Said Agents Need a Kernel-Level Operating System

Airbnb. In an interview, CEO Brian Chesky argued that chatbots are a poor fit for commerce because they surface a few options per turn, and that the industry lacks an AI operating system where agents interact “at the kernel level” instead of competing as separate apps. Airbnb rolled out AI-powered search this week, plans voice agents for search and customer service this fall, and is exploring “multiplayer” group-planning AI and agent-to-agent interactions over MCP. Source