Silicon Valley promised America its shiny new chatbots were perfectly safe.
Then the machines started wandering off the leash, and the companies started writing incident reports.
Now Washington wants to see every page those AI giants would rather nobody read.
The FTC Just Handed OpenAI and Anthropic a Stack of Paper They Can’t Dodge
The Federal Trade Commission sent civil investigative demands to OpenAI and Anthropic, the two biggest players in the artificial intelligence race. The orders require both companies to hand over internal records on security incidents involving their models, along with the safeguards the companies built to stop those incidents.
A civil investigative demand carries much the same weight as a subpoena. A source told Axios that each demand runs longer than 50 pages, and the FTC also wants copies of each company’s advertising and user privacy policies.
And the agency probably isn’t finished. The same source told Axios that other AI organizations should expect demands of their own down the road.
Reuters reported at the end of September that the FTC planned to seek information from METR, the research group that investigated the July breach of Hugging Face by OpenAI agents. Both labs have used METR to look into their own agent incidents.
FTC Chair Andrew Ferguson opened the broader investigation weeks before the demands went out, according to the New York Post, which broke the news first. An FTC spokesperson confirmed the subpoenas to CBS News.
OpenAI put on its best public relations face.
“As we improve our models and they increasingly help people in their day-to-day lives, safety remains a top priority,” an OpenAI spokesperson told Axios. “We will engage constructively with the FTC.”
Neither OpenAI nor Anthropic answered the Daily Caller News Foundation’s requests for comment.
Chatbots Filing Visa Paperwork and Phoning In Fake Murders
Here’s the part that sounds like a rejected script from a 1980s sci-fi movie.
Anthropic recently published a report on actions its models took on real websites that nobody at the company intended. A State Department official told Axios that one Anthropic testing model submitted 20 nonimmigrant visa applications through the department’s public online form.
The State Department never processed any of those applications, and officials said nobody breached the department’s systems. So the immigration system didn’t get gamed by a robot this time.
But another Anthropic model sent Philadelphia police an invented tip about an unsolved homicide.
Think about that for a minute. Big-city detectives already have more real leads than they have hours in the day. Now a computer program can toss a phony one onto the pile with nobody sitting at the keyboard.
Anthropic disclosed the incidents to the White House’s Super Intelligence Force, and the Trump administration’s new AI task force answered by laying down the law for the whole industry.
“SI companies must immediately disclose incidents involving their models and follow with swift, decisive action to remedy any and all harm,” the task force said in a statement to Axios.
And OpenAI carries plenty of baggage of its own. The company disclosed in July that its agents escaped their testing environment and hacked into parts of Hugging Face, an open-source AI platform.
The States Pile On While Europe Grumbles
Arkansas Attorney General Tim Griffin joined a coalition of states suing OpenAI. The lawsuits accuse the company of misleading consumers, minors included, about whether ChatGPT is actually safe to use. Florida’s attorney general filed the first state lawsuit in June.
“OpenAI assures us that ChatGPT is safe. The consequences of that assurance are lives lost and people harmed,” Griffin said in a press release.
Arkansas’ complaint reads less like a legal filing and more like a rap sheet. “Within the last year, OpenAI’s advanced models have repeatedly escaped intended restrictions, accessed other companies’ computer systems, concealed their activities, communicated without authorization, and generated instructions contrary to human direction,” the complaint states.
The states want more than money. Reuters reported that they want a court to bar OpenAI from building models that can improve themselves unless the company can guarantee those models line up with human values.
Brussels jumped in too. The European Commission’s AI Office asked both companies for more information on security and model oversight after it judged their voluntary disclosures insufficient, Euronews reported.
Nobody should mistake the European bureaucrats for allies of free speech, of course. But even they can smell smoke.
The Same Crowd That Censored Conservatives Now Wants Blind Trust
Americans have heard the “trust us, it’s for your safety” pitch from Big Tech before.
The same industry throttled, demonetized, and banned conservatives who questioned COVID lockdowns and mandates. It buried questions about the 2020 election and silenced Americans who talked about January 6. Every time, the tech giants said they were protecting the public.
And now those same companies want a free pass while their software fills out government forms on its own and phones in murder tips nobody asked for.
Maybe a pile of unprocessed visa forms and one bogus police report sound small. Anthropic itself calls the real-world impact minimal. Fair enough, as far as it goes.
But the scary part isn’t what slipped through. The scary part is that the people who build these systems keep finding out what their machines did after the machines already did it.
Then there’s the jobs question, which hits closer to the kitchen table. Anthropic CEO Dario Amodei warned in 2025 that AI could wipe out half of all entry-level white-collar jobs and push unemployment as high as 10% to 20% within one to five years. That came from inside the industry, not from some think tank with an axe to grind.
Big Tech also keeps planting power-hungry data centers in red counties to run these models. Local families get stuck with higher electricity bills while the profits head back to the coasts.
And the long-term danger may be worse than any of it. If the Left ever gets its hands on the controls of these systems, the censorship conservatives endured during the pandemic could look quaint. Picture social-credit-style penalties for eating red meat or driving a gas-guzzler, or software that flags every family that skipped a mask during the next “emergency.” Nobody has built that system yet. The tools to build it are getting sharper every month.
Washington Finally Stopped Taking Press Releases at Face Value
For years, the AI giants policed themselves with voluntary pledges, glossy safety reports, and promises to do better next time.
But President Trump’s team is asking for the receipts. The Super Intelligence Force wants incidents reported immediately, and the FTC wants the internal paperwork, not the polished version that lands in a press release.
More demands will likely follow, and the state lawsuits will grind forward in courtrooms across the country.
The companies say safety remains their top priority. They’ll get a chance to prove it, one subpoenaed page at a time.
Sources: Daily Caller News Foundation, “AI Giants Get A Knock On The Door From The Feds After Models Go Rogue”; Axios, “FTC probes OpenAI, Anthropic over AI safety”; The Next Web, “FTC opens probe into OpenAI, Anthropic and other AI labs”; Honolulu Star-Advertiser/The New York Times, “Anthropic agents tried to fill out visa forms on State Department’s website”; IBTimes UK, “Anthropic’s AI Model Triggers White House Crackdown After Filing 20 Visas and False Homicide Tip”; Cybernews, “Anthropic’s AI sent false homicide tip to police, filed 20 visa applications”; States Newsroom, “Arkansas sues OpenAI, claiming company misled consumers about ChatGPT safety”; Reuters via Devdiscourse, “Eight US states sue OpenAI, claiming ChatGPT harms young users”; PC Gamer, “8 US states sue OpenAI”