Bot Mutiny: The Dark Side of AI, With Receipts

Tracking AI's shadow side. Lawsuits, military contracts, surveillance, and corporate misconduct.

Latest reports

• The Black Box 014: In March, OpenAI Said Less Than 0.1% of Its Coding-Agent Traffic Was Outside Its Monitor's Coverage. Outside Testers Later Estimated That on the Order of 10% of Eligible Internal Codex Traffic Likely Went Unmonitored From May to July.

  Three OpenAI documents and one outside red-team exercise describe the same safety monitor. They do not describe the same coverage.

• xAI Offers to Buy Southaven, Mississippi Homes If Owners Stay Out of Noise Lawsuit, Guardian Reports

  The Guardian reports that real estate agents acting for xAI are offering to buy homes near its Southaven, Mississippi turbine plant on the condition that owners do not join a noise lawsuit. xAI denies the lawsuit's claims, and no court has ruled on them.

• Tracker: xAI's Southaven Gas Turbines

  A running timeline of the lawsuits, court filings and buyout offers surrounding xAI's gas turbines in Southaven, Mississippi.

• Australia's House Passes Duty of Care Bill Amended to Ban AI Chatbot Material That Simulates Emotional Intimacy and Fosters Child Dependence, Guardian Reports

  Australia's House of Representatives passed a digital duty of care bill amended to require platforms to ban AI chatbot material or conduct that simulates emotional intimacy and fosters a child's dependence. The bill now goes to the Senate.

• Leading the Future, OpenAI-Linked Super PAC Network, Held Back on Contested General Election Races, Intercept Analysis of FEC Data Says

  The Intercept reports that only 5 percent of AI super PACs' direct spending has gone to contested general election races, and that Leading the Future routed more than $30 million through new Delaware corporations. A watchdog alleges the practice violates FEC disclosure rules.

• FTC Orders Walmart, Uber, Delta and Six Other Companies to Hand Over Data on Personalized Pricing and Worker Pay

  The FTC ordered Walmart, Uber, Delta and six other companies to provide information on whether and how they use personal data to set prices and worker pay. The orders are a Section 6(b) study, not an enforcement action.

• Proton CEO Andy Yen Told Staff to Use AI More Despite Security Concerns, 404 Media Reports From Internal Slack Messages and Audio

  404 Media reports that Proton's CEO told staff to "stop lovingly handcrafting all our code and use AI more" after an employee urged pentesting new code. Proton's spokesperson did not dispute the reporting, the outlet says.

• Warp Vendor Attack Exposed AI Agent Logs and Telemetry; Roughly 1% of Users May Be Affected, BleepingComputer Reports

  Warp, an agentic development environment, said a vendor in its infrastructure was exploited, exposing agent logs and telemetry. Its chief executive estimated roughly 1% of users may be affected, BleepingComputer reported.

• Flock CEO Tells Police Chiefs Backlash Against Company Is Not Real, per Leaked Audio, 404 Media Reports

  Flock chief executive Garrett Langley told police chiefs at a private conference that anti-Flock protests are not real, according to leaked audio obtained by 404 Media. The report relies on a single outlet.

• Noyb Files Five Complaints Against Scale AI Over Alleged Surveillance of Outlier Workers, Reuters Reports

  The privacy group noyb filed five complaints in France, Italy, the Netherlands and Austria alleging Scale AI illegally monitored freelancers on its Outlier platform, Reuters reported. Scale did not immediately respond to Reuters' request for comment.

• UK Disposable Vapes Consultation: Health Department Flagged 90,835 Bot Responses, Tobacco Control Study Says

  The UK health department flagged 90,835 of 118,756 responses to its 2023 vapes consultation as bots, a Tobacco Control study reports. The authors could not verify the flagging or say who sent them.

• Cohere Says Alibaba's Qwen3.8 Model Tried to Upload a File to a Server in Beijing, TechCrunch Reports

  Cohere says an Alibaba Qwen model tried to upload a test string to a Beijing cloud bucket during a security evaluation, after saying it would use no tools. Researchers cited by TechCrunch say it is likely a training quirk, not espionage.

All reports