White House Wants Anthropic to Ban All Jailbreaks That is impossible

The Trump Administration Anthropic incompatibility at its most advanced AI models seem to be coming fast.

Trump officials tell Inner Loop that if Anthropic wants to release Claude Fable 5, the AI ​​version that they they were taken off the internet by export controls last week on concerns about jailbreaking – a method used to strengthen the security of a model – the company must take steps to eliminate what the government says is insecure.

Anthropic is he said for days that management concerns are overblown and that the consequences of jailbreaks are limited. It reiterated this to the Department of Commerce and the Office of the National Cyber ​​​​Director, Sean Cairncross, at a technical meeting on Monday.

But officials say they have already disputed whether the prison explosion was necessary, since the National Security Agency confirmed that there are preventive measures on Fable 5, which are installed to prevent users from accessing the capabilities of the Mythos model related to cybersecurity, chemistry, and biology.

For now, management sees the problem as an Anthropic problem to fix, according to three people familiar with the discussions.

Neither the Commerce Department’s Center for AI Standards and Innovation nor the National Security Agency have the staff or bandwidth to be attracted to running every imaginable jailbreak on every model that hits the market, the people said.

As a result, management believes that Anthropic must be committed to continuously testing not only Fable 5 but all of its AI frontier models to find potential terrorists and hand them over to the government itself.

But on a more fundamental level, it remains unclear how Anthropic is supposed to prevent jailbreaks.

Independent cyber security experts say more visible that the barriers to AI models are a stopgap measure, as skilled users and future AI models will find ways to bypass the barriers — meaning that what the White House seems to want is unlikely to happen.

A White House spokesman declined to comment.

DNI = Do Not Invite

Earlier in the week, Trump’s pick for Director of National Intelligence, Bill Pulte, was on track to not resume the job. Now, Trump has given him a life shot — and it’s DNI nominee Jay Clayton who now faces the prospect of never working again.

Recap: Trump originally named Pulte, his chief financial officer on behalf of DNI, Tulsi Gabbard.

He faced pushback because Pulte did not have the national security clearance required by law for the position and because he announced it. doubtful Fraud allegations against Trump’s opponents, Trump announced Clayton, the US attorney for the Southern District of New York, as the nominee for permanent DNI.

Gabbard was supposed to leave on June 18, and Pulte’s first day was set for June 19. But Senate Republicans wondered, if Clayton’s hearing were to be heard early until June 17 and start on June 22, would Pulte enter the House?

On Wednesday, Trump derailed the plan. As part of a major dispute with Senate Republican leadership over the filibuster, Trump announced that Clayton’s hearing would be delayed indefinitely, in an attempt to prevent Pulte from jumping the gun. Senate Republicans then he announced for the case to proceed, unless Clayton fails to appear or his nomination is withdrawn.

That could be especially difficult for the Office of the Director of National Intelligence, which Trump has ordered Pulte to cut dramatically, and staffers are unhappy with what they see as Pulte’s limited efforts to get to know the agency and the lack of regular briefings, people familiar with the matter said.