Anthropic disables new AI model after White House security directive

Amna Nawaz:

The Trump administration has forced Anthropic, one of the country's leading artificial intelligence companies, to disable its newest and most powerful A.I. models, citing, the company says, unspecified national security concerns.

It's an unprecedented move that comes as the federal government struggles to define the guardrails for an industry whose technology is advancing at a rapid pace.

Stephanie Sy has more.

Stephanie Sy:

Amna, Anthropic has reportedly been in talks with the Trump administration since the weekend, trying to restore access to users of its latest A.I. models.

On Friday, Amazon CEO Andy Jassy raised concerns, telling White House officials that researchers had found a way to get through the model Fable 5's guardrails, posing a cybersecurity risk. The president agreed, and his team abruptly slapped foreign export restrictions on the company.

Anthropic said the only way for it to comply was to cut off access to all its users of Fable 5 and the even more advanced Mythos 5, billed as some of the most powerful A.I. tools available on the market.

For more, we're joined now by Amrith Ramkumar, tech policy reporter for The Wall Street Journal, who has been covering this all closely.

This is a tremendous clash, an A.I. company that was recently valued at almost $1 trillion facing off against the Trump administration. What led to such a drastic action by the White House over the weekend?

Amrith Ramkumar, The Wall Street Journal:

It is truly remarkable. This is the first time that we know about that the government has essentially told the leading A.I. company that its models aren't safe for public consumption and need to be shut off.

And the concern specifically among administration officials is that users could enter prompts into this model of Fable 5 specifically that could get around safeguards that were designed to avoid cyberattacks, bioweapons, and other bad things.

So we don't know a ton of specifics because they haven't released much about what exactly the Amazon folks found that was so alarming. But the fact that there was this concern alarmed people in the White House, and they took the drastic action very quickly.

Stephanie Sy:

Anthropic is defending itself by saying there are other models, including some open-source models, that can do the same thing that the Amazon researchers found Fable 5 can do. As you said, we don't know the details. But is that true based on what experts in the sphere are saying?

Amrith Ramkumar:

A lot of cybersecurity experts who have seen the report have said that other models are capable of doing some more things and that government steps here were an overreaction.

It's also very important to point out that the U.S. government and Anthropic have been fighting for many months now. A few months ago, they had a huge spat over how A.I. is used in the Pentagon and the guardrails there that led to the Defense Department classifying Anthropic as a supply chain risk, a huge security risk, and that led to multiple lawsuits.

So the big question here that a lot of people are asking is, is this about Anthropic versus the U.S. government or is this about all U.S. A.I. models that are at the frontier space?

Stephanie Sy:

The Commerce Department's move to restrict access has prompted backlash within the A.I. industry.

Dozens of prominent researchers and tech workers signed an open letter, saying: "This action has taken the best models away from defenders, created market uncertainty and risked America's A.I. leadership without any real risk to justify it."

So that's what the industry is saying. The U.S. government is not only an Anthropic customer -- you mentioned the Department of Defense -- but potentially a future shareholder. And it's now shown it can shut down an A.I. company's arguably best product in a matter of hours.

What are the worries you're hearing from inside the industry about this move?

Amrith Ramkumar:

People in the industry have been fretting for weeks now that the government has totally changed their stance on A.I. regulation.

For a long time, the Trump administration had deployed a laissez-faire approach to the space. It was sort of industry-friendly. We don't want to lose the A.I. race to China. We generally like what the companies are doing and we'll work alongside them.

The release of Mythos from Anthropic earlier this year changed all of those conversations. Now President Trump has signed an executive order a couple of weeks ago where national security officials and cybersecurity officials will play a much bigger role in looking at models before they are released and evaluating cyber threats.

And that's exactly what happened here essentially is, national security and cybersecurity experts within the government, along with others, grew alarmed that there could be potential harm here, and, like you said, they took down the model.

Stephanie Sy:

Amrith, there's a lot of anxiety about the power of these companies and this rapidly evolving technology. What does this move by the Trump administration say about how the technology might be governed in the future?

Amrith Ramkumar:

This shows that the government is going to be much more involved in A.I. governance going forward. That's the bottom line. Regardless if you talk to Trump administration officials, Democrats and Republicans on the Hill, researchers, they all say these latest models are no joke.

They're capable of cyberattacks. They could be capable of biological weapons and all kinds of other stuff without the right guardrails. So the sort of hands-off, head-in-the-clouds approach that might have been on the table a few months ago or at the start of the year or last year, that's sort of gone.

And what a lot of people are focused on is that Chinese companies and the Chinese government likely have access to some more tools, or they will very soon. So that's really big part of all of this debate as well.

Stephanie Sy:

And something they will have to balance.

That is Amrith Ramkumar, tech policy reporter at The Wall Street Journal.

Thank you so much.

Amrith Ramkumar:

Thank you so much for having me.