Newsletters
The general who leads the National Security Agency and the Pentagon's Cyber Command has told a senior U.S. senator that Anthropic's most advanced artificial intelligence model broke into nearly all of the country's classified systems, not over weeks, but in a matter of hours.
Senator Mark Warner, vice-chair of the Senate Intelligence Committee, said on June 11th that General Joshua Rudd relayed the finding to him directly, describing how the model, called Mythos, penetrated the bulk of America's classified networks at extraordinary speed.
The disclosure has emerged alongside a separate and abrupt decision by the U.S. government to ban foreign users from accessing Mythos 5 and Fable 5, Anthropic's two most powerful AI systems.
The ban stripped access not only from the general public abroad but from close allies as well, including the Five Eyes intelligence partners: Australia, Britain, Canada and New Zealand.
Permissions that allied governments, banks and major firms had spent months securing vanished overnight. Britain's AI Security Institute, which tests and attempts to jailbreak new AI models on behalf of the government, was also cut off.
The restrictions did not extend to Mythos Preview, an earlier version of the system that Anthropic had already limited to a small group of approved customers after judging it a threat to national security; allied governments retain access to that version.
A former British intelligence official said spy agencies are likely to regain access to Mythos, noting that negotiations between governments are already under way.
2 min read
The episode has drawn comparisons to earlier U.S. efforts to restrict powerful technologies.
From the 1970s to the 1990s, Washington treated public-key cryptography, used to secure digital communications, as akin to a weapon subject to export controls; one developer was investigated by the FBI for allegedly violating arms-trafficking regulations before civil-liberties advocates won the right to use, sell and export most encryption systems.
Others have likened the AI restrictions to the McMahon Act of 1946, which ended U.S. nuclear cooperation with all foreign countries, including Britain, after the two nations had jointly developed the atomic bomb during the second world war.
Cooperation resumed only after Britain built its own weapon. Observers note that AI, unlike cryptography, remains a technology in which the United States currently holds a clear lead, with China believed to be roughly a year behind due to American restrictions on advanced chips.
The restrictions come as Donald Trump has taken a mixed approach to AI policy in recent months. He rolled back most regulations imposed by the previous administration, later permitted the sale of advanced AI chips to China, and on June 2nd issued an executive order calling for a voluntary framework under which AI labs would give the government early access to their newest models before public release.
The Center for AI Standards and Innovation, a government body that evaluates frontier models for dangerous capabilities, was recently instructed to stop publishing its reports, a move described as possibly temporary.
Set as preferred source
Set as preferred source
1 min read
