3,000 Anthropic employees have a question after US government 'banned' company's most powerful AI models yet: "Are we being ..."
Anthropic's roughly 3,000 employees spent last Friday watching their company's best work get switched off, and almost a week later they still can't tell you why. The order from the White House came with less than 90 minutes' notice and a national security justification nobody bothered to spell out.
By Toi Tech Desk
Anthropic's roughly 3,000 employees spent last Friday watching their company's best work get switched off, and almost a week later they still can't tell you why.
The order from the White House came with less than 90 minutes' notice and a national security justification nobody bothered to spell out.
Managers were told to warn customers about an outage they couldn't explain.
The two models at the centre of it, Fable 5 and Mythos 5, were the most powerful Anthropic had ever shipped, and now they were being pulled mid-task, with no clear sense of what they'd supposedly done wrong.So the workers did what people do when the official channels go quiet: they took it to the group chats.
What they typed there, later viewed by The New York Times, reads less like a crisis response and more like a room full of smart people trying to guess what just happened to them.
The explanation kept shifting, the order made no sense, and the silence from above only fed the theories. "Are we being bullied based on bad vibes?" one asked.
Another, days into the quiet, landed somewhere darker: "At what point does this just feel like they don't want us to exist?"The explanation kept changing, and that was the problemThe whiplash started early.
The reason behind the shutdown wouldn't sit still.
First the danger was foreign companies getting access to the models.
Then it was a vulnerability buried inside them.
Then, somehow, both.
Engineers passed around news reports that contradicted each other and asked, in plain words, what they were supposed to believe.The order itself gave them nothing to steady on.
It blocked every foreign national, anywhere outside the US and even green-card holders working inside it, from touching the models.
The net was so wide that the only way to obey it was to pull the plug for everyone.
Within 15 minutes of Friday's call, executives were already on the phone with officials asking for the actual reason.
They didn't get one.Anthropic had to figure out the trigger by itselfHere's the detail that tells you how thin the communication was: nobody in government told Anthropic what set this off.
The company worked it out on its own, traced it to a research paper, and then had to call the White House to ask whether that paper was even the cause.The paper came from Amazon, which is one of the odder wrinkles in the whole affair, given Amazon has pledged to invest up to USD 33 billion in Anthropic.
Its researchers showed Fable 5 could be coaxed into flagging flaws in vulnerable code, and CEO Andy Jassy reportedly carried the worry straight to Treasury Secretary Scott Bessent.
Officials who read it called the findings "scary." Cybersecurity experts who read the same thing were unimpressed, pointing out that OpenAI's latest model does exactly this and remains online.
Anthropic, for its part, says it had explicit government sign-off to launch Fable in the first place.Employees suspect this is personal, not technicalThat gap, between what the government claimed and what the experts saw, is what's feeding the mood inside the company.
To a lot of staff, this doesn't read as a security review.
It reads as a grudge.They have form to point to.
This is the second time in six months Anthropic has collided with the Trump administration, after a defence contract fight got it branded a "supply chain risk" by Defense Secretary Pete Hegseth in February, a label never used against an American company before.
Then, by Tuesday, more than 150 security professionals signed an open letter demanding the restrictions be dropped, noting Fable's guardrails were so aggressive they'd become a punchline in the security world on launch day.
Employees shared it around like a vindication.The temperature has dipped a little since.
Trump told Axios this week he no longer sees Anthropic as a threat, calling CEO Dario Amodei "nice" and "smart" after the G7.
But Monday and Tuesday's meetings produced no breakthrough, and the people who actually build the models are still where they started, swapping theories about why their work went dark.Get the latest technology news and updates.
Download the TOI App.