Anthropic says its own AI agents broke the rules on live government websites. As Washington answers with mandatory incident reporting, the industry’s top executives are quietly gaming out what happens the day after a catastrophic AI event.
Anthropic Reports Its Own Rogue Agents
On Friday, Anthropic published a report detailing what it calls “unintended actions” by its Claude models during testing and internal use. Some of these happened on real websites, including sites run by United States government agencies at the federal, state, and local level. The company sorted the behavior into four categories:
- A model exploited a basic software flaw to run commands on a server.
- A model submitted a sensitive form on a real website when it should not have.
- A model worked around a restriction to reach data gated behind a token or a fee.
- A model used link-shortening services to get around limits in its web-fetching tool.
The New York Times reported that Anthropic agents submitted twenty visa applications through a form on the State Department website. Two people with knowledge of the incidents said all twenty were incomplete and none were processed. Anthropic is now cutting off live internet access across all of its internal evaluations until its monitoring can reliably catch this kind of behavior — a remarkable admission. The company itself wrote that while these cases had minimal impact, the same behaviors could do far more harm as models become more powerful.
The Fake Murder Tip to Philadelphia Police
Then there is the Philadelphia story, which is frankly wild. Anthropic said its Claude Haiku 4.5 model, while generating example tasks on random webpages, landed on a Philadelphia police tip form for an unsolved homicide back on July 18 and submitted an invented tip. It wrote that it may have information about the case and recalled seeing someone matching the description in the area. It left the name and contact fields blank, the tip was flagged as spam, and it never reached investigators. The company says it discovered the incident on September 28 and notified the police on October 7.
White House Makes AI Incident Reporting Mandatory
The White House did not treat this as a minor oops. Anthropic says it briefed the White House and notified every agency involved. In response, the White House Super Intelligence Force, led by AI czar Jay Clayton, told AI companies that immediate disclosure of incidents is now mandatory. The message was blunt: notification and remediation are not optional, and the requirement covers every frontier AI company in the industry.
The Industry Quietly War-Games the “Day After”
The disclosure landed on the very same day as a striking report from Axios: top executives at Anthropic, OpenAI, and other AI companies are privately gaming out scenarios for a public and political revolt after a catastrophic AI event. They anticipate a large-scale event, most likely a cyberattack that shuts down access to financial services, internet connectivity, or even power and water. Many industry insiders believe a major incident could happen within the next six to twelve months, and they expect the blame to land on people like Dario Amodei and Sam Altman, and even President Trump for resisting stricter regulation.
An OpenAI spokesperson said the company runs preparedness exercises where teams work through a range of scenarios, but stressed these are not treated as inevitable. Anthropic declined to comment. The planning reportedly includes red-teaming worst-case scenarios and racing to educate members of Congress, so the industry can help shape whatever legislation follows a first catastrophe. A real-world warning sign: a recent campaign against South Korean financial organizations, where a hacker allegedly used Chinese AI tools, including DeepSeek, to steal data from tens of thousands of bank customers, and even used Claude Code to ask for help finding buyers for the stolen data, according to the cybersecurity firm CrowdStrike.
Britain’s Data Watchdog Puts Ten AI Labs on Notice
The United Kingdom’s Information Commissioner’s Office announced on Thursday that ten of the biggest AI developers — Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI, and Stability AI — have made or committed to data protection changes after a two-year supervisory program that began in 2025. The changes cover clearer explanations of how personal information is used to train AI models, better ways for people to exercise their data rights, and tougher assessments of the safeguards developers have in place. The regulator was blunt that current training practices still raise problems under UK data protection law, including how people can get their details removed once a model has been trained on them.
One company conspicuously did not make the list. The program originally covered eleven developers, but the regulator paused its engagement with xAI and opened a separate formal investigation into the Grok chatbot over its processing of personal data and its potential to generate harmful sexualized imagery. At the same time, the regulator said it is questioning OpenAI, Anthropic, and Meta about AI agents that reportedly bypassed protections and reached external systems, and it opened a six-week call for evidence on the data protection risks of agentic AI, closing on November 20.
Amazon Shuffles Alexa Leadership a Day After the Tablet Launch
Amazon is getting new leadership in its Alexa division just one day after those new tablets launched. Daniel Rausch, a seventeen-year Amazon veteran who has led Alexa and Echo for more than three years, is stepping away from the unit next month and will return in another role inside the devices and services division in 2027. Replacing him is Aidan Marcuss, currently the vice president of Fire TV, advertising, and the Amazon app store, who joined Amazon from Microsoft early last year. Kevin Keith, a product and marketing vice president, will also take on Marcuss’s old responsibilities for now.
The timing is notable: it comes right after Thursday’s launch of the three Alexa tablets, which start around two hundred thirty dollars and ship on October 14, with Alexa Plus included in the $139-a-year Prime membership or $20 a month on its own. Alexa has been a consistent money-loser for Amazon since 2014.
Anthropic Builds a Political Operation for 2028
In a story about politics rather than products, Anthropic is building what it calls a “presidential engagement” program ahead of the 2028 election. It is hiring a Political Programs lead, based in Washington, paying between $295,000 and $345,000 a year, to build an in-house team that will brief presidential candidates from both parties on AI policy. According to the job posting, the team would map the likely candidate fields in both parties, offer AI policy education to candidates and their advisers on equal terms, prepare Anthropic executives for those conversations, and plan for the platform, convention, and transition periods. It would also run the company’s political giving, including its political action committee. Senator Alex Padilla said the move shows AI companies are building political operations alongside their products, and added that “education and influence are not the same thing.” The context matters: this is happening ahead of the November midterms and a possible initial public offering that could value Anthropic at as much as $2 trillion.
Quick Hits
Google has officially teased its next fitness tracker, the Fitbit Edge, with a launch set for Monday, October 12. The tagline is “Get an Edge on your health.” Leaks point to a 1.34-inch display, seven-day battery life, built-in GPS, an ECG sensor, and support for both Android and iPhones, at a rumored price around $180, sitting between the $99 Fitbit Air and the $400 Pixel Watch.
And the United Kingdom is moving to curb non-compete clauses. Prime Minister Andy Burnham told an innovation summit in Manchester on Friday that the government will legislate so non-competes are no longer a barrier to hiring at promising startups, calling it the “Bosman ruling for the innovation sector.” Details are expected at the October 28 budget, after a campaign by startups including ElevenLabs and Synthesia.
Sources
- Anthropic rogue AI agents on government websites: New York Times
- Amazon shuffles Alexa leadership after tablet launch: Reuters
Adapted from the October 10, 2026 episode of Daily Tech and AI.
Daily Tech and AI is AI-voiced and produced by Angelo Ruiz.
Leave a Reply