Latest / HotTea: The AI Morning Briefing / Cybercab hit public roads, then the audit arrived
Transcript
- Sections 00:00 What we're covering today 00:25 1. Federal regulators opened a Cybercab audit hours after its public debut 02:06 2. 128 states agreed on a path toward autonomous-weapons rules, but not on the rules 03:08 3. A judge kept Minnesota's nudification ban alive, and xAI will appeal 04:13 4. A trick built to fool AI agents spread into millions of phishing emails 05:17 5. DeepSeek made a 75 percent model discount permanent 06:07 Visit Hot Tea Disclosure Narration uses an AI-generated voice. Transcript Good morning. It's Sunday, September 6, 2026. This is HotTea. Tesla put a robotaxi without a steering wheel or pedals on public roads. Hours later, federal safety officials opened an audit into how the company certified it. The question is not whether a driverless car needs a steering wheel. It is what proof replaces the rules built around one. Federal regulators opened a Cybercab audit hours after its public debut. Tesla began giving members of the public rides in its two-seat Cybercab in Austin. The vehicle has no steering wheel or brake pedals. It also lacks some other equipment that federal safety standards normally require. Hours after the launch event, the National Highway Traffic Safety Administration opened an Audit Query into Tesla's certification process. U S. automakers certify their own compliance with Federal Motor Vehicle Safety Standards. NHTSA said its query would examine the process and technical data Tesla used. The agency also said it would review whether Tesla decided that some standards did not apply to the Cybercab. Tesla has said it does not need an exemption because the vehicle complies with current rules. The Verge reported that the query covers an estimated 1,000 Cybercabs. Texas records listed 45 in Tesla's fleet. Tesla plans to use the cars in its Robotaxi service in Texas and Florida. The company did not answer requests for comment from Ars Technica and Wired. The pressure: An Audit Query is an investigation, not a finding that Tesla broke a rule or built an unsafe vehicle. NHTSA has not ordered a stop to Cybercab rides. The estimated fleet size comes from reporting; Texas records showed 45 registered vehicles. What to watch: NHTSA can request Tesla's test data, certification logic and production records. The agency can require changes or pursue penalties if it finds noncompliance. The first concrete test is whether Tesla can keep charging for public rides while the audit continues. Zoox offers a comparison. It sought an exemption for a vehicle without normal controls and received one this summer, limited to 2,500 vehicles this year. Watch for signs that NHTSA tells Tesla to take the same route, narrows the Cybercab's service, or accepts Tesla's reading of the standards. 128 states agreed on a path toward autonomous-weapons rules, but not on the rules. The 128 members of the Convention on Certain Conventional Weapons reached consensus on a non-binding document in Geneva. Reuters reported that the text could open a path to formal treaty talks after more than a decade of debate. The weapons under discussion can identify and engage targets with varying degrees of autonomy. The United States and Russia have favored national guidelines over binding international rules. Consensus does not create a treaty or restrict a weapon today. Stop Killer Robots told Reuters that negotiators diluted the definition and civilian protections in the final hours. The official UN meeting record confirms the 2026 process. Reuters supplies the reported overnight outcome and the dispute over the text. The next test is whether states authorize formal negotiations and publish a definition that covers the systems now used in war. Watch the language on human judgment, target selection and civilian harm. It will show whether the process sets a real limit or only records that many states are concerned. A judge kept Minnesota's nudification ban alive, and xAI will appeal. U.S. District Judge Donovan Frank denied xAI's request for a preliminary injunction. Minnesota's law therefore remains in effect while xAI's First Amendment challenge proceeds. The law bars website operators, software developers and others from letting users create realistic fake nude images of identifiable people. It took effect on August 1. xAI said in a court filing that it would appeal to the Eighth Circuit. The ruling does not settle whether the law is constitutional. Frank said the issues are complex and deserve full review. At this stage, he found that xAI had not shown the harm needed for an injunction. Minnesota's law regulates access to a category of tool. That means the final case can reach beyond a rule aimed only at distributing nonconsensual images. The Eighth Circuit can pause the law or leave it in force. Watch how the courts treat the law's scope, its technical-skill exemption and its penalties. Other states will decide whether to copy Minnesota's approach to tool access or use narrower consent and takedown rules. A trick built to fool A I agents spread into millions of phishing emails. Microsoft researchers found a phishing campaign using invisible Unicode tag characters inside finance words. A recipient still saw words such as "funding," while some filters saw broken-up tokens. Microsoft built the detector while studying hidden prompt injection in email. Its signature jumped from about 21,000 messages on February 8 to more than 1.3 million the next day. It later peaked at 2.37 million messages on one weekday. The pressure: The counts come from Microsoft Defender telemetry, so they are a company claim about one vendor's view of the campaign. Microsoft said layered defenses caught most messages. The technique did not invent invisible-character spam. The new part was the Unicode Tags block, which security researchers had recently made prominent through A I red-team work. What to watch: Email providers should say whether they normalize characters from U+E0000 to U+E007F before tokenization and filtering. Defenders also need to avoid false positives from the England, Scotland and Wales flag emoji sequences, which legitimately use tag characters. DeepSeek made a 75 percent model discount permanent. DeepSeek will keep prices for its flagship V4-Pro model at one quarter of their original level. The company said that in a statement reported by Reuters on Saturday. That turns a temporary discount into the normal price. DeepSeek's API pricing page says prices vary by peak and off-peak hours and warns that product prices can change. The permanence and 75 percent figure are company claims reported by Reuters. A posted token price does not show service capacity, model quality or the full cost of using the system. DeepSeek's pricing page also reserves the right to adjust prices again. Rival API price changes will show whether this starts another price round. Usage growth, rate limits and service reliability will show whether the lower price creates more durable demand or shifts costs into capacity limits. That is the signal before the noise. This briefing was produced from Hot Tea's verified daily edition. For the complete briefing and every source link, visit Hot Tea dot A I.