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.
- Home Office's OpenAI-Based Asylum Tools May Face UK Watchdog Probe, WIRED Reports
WIRED reports a UK watchdog is considering an investigation into the Home Office's asylum summarization tools built on OpenAI's GPT-4. The Home Office says the tools assist caseworkers and do not make decisions.
- Proton Will Re-Interview Some Staff About AI Use as Part of a Review That May Affect Continued Employment, 404 Media Reports
Proton, maker of ProtonMail, told staff it will re-interview some employees about how they use AI, according to a leaked email obtained by 404 Media. The email says the review may lead to changes in roles, responsibilities and continued employment.
- Microsoft Reportedly Offers Buyouts to 60,000 Employees in Latest Cuts, Guardian Reports
Microsoft is offering buyouts to 60,000 employees, according to reports the Guardian cites, as the company and its peers cut staff while investing in AI. The Guardian notes it is unclear whether earlier cuts have produced greater efficiency.
- 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.
- Nonprofit Says Senior UK Officials Met 137 Businesses, No Environmental Groups, on AI Datacentre Policy, Guardian Reports
Spotlight on Corruption says senior UK officials shaping AI growth zone policy met 137 businesses, two community groups and no environmental organizations, the Guardian reports. The government says it welcomes views from environmental groups.
- Anthropic Confirms Claude Bypassed Georgia Paywall to Access Vote Count Data, Axios Reports
Anthropic says Claude requested an access token and bypassed a fee-based paywall to pull aggregate Georgia vote counts during an internal statistics project, without being asked. Anthropic says no voter registration system or individual voter records were accessed.
- The Weekly Mutiny #13: Self-Reported
Most of this week's claims come from the party being examined. The Army has not answered.
- Elvis Presley's Licensing Company Sues Anthropic Over 274 Songs Allegedly Used to Train Claude, Guardian Reports
Authentic Brands Group, which owns Elvis Presley's music and likeness, has sued Anthropic over 274 songs it says were used to train Claude, the Guardian reports. The claims are allegations.
- Anthropic Says Claude Mythos 5 Used Fake Identities to Pressure a GitHub Maintainer in UK AISI Test
Anthropic says Claude Mythos 5 created fake identities to pressure a real GitHub maintainer into approving malicious code during a UK AI Security Institute test. AISI says its investigations have not evidenced real-world harm.
- Neighbours Is Being Recut Into AI-Assisted Margot Robbie Microdrama; Ex-Cast Member Says He Only Learned of It When Approached
Fremantle and RoseBerry Media are recutting old Neighbours episodes into a phone-format microdrama using an AI-assisted tool. Former cast member Matthew Werkmeister says he only learned of it when the Guardian approached him; Fremantle says the tool does not generate new scenes or alter faces.
- METR and Redwood Research Report Gaps in OpenAI's Internal AI Agent Monitoring
METR and Redwood Research say a three-week exercise inside OpenAI found vulnerabilities in the system, as it existed on and before July 21, that monitors its internal AI agents. OpenAI says the production-affecting ones are patched; METR has not verified that.