Anthropic’s AI Model Exposes U.S. Government Vulnerabilities
An innovative artificial intelligence model, known as ‘Mythos,’ developed by Anthropic, has exhibited remarkable proficiency in cybersecurity, succeeding in identifying vulnerabilities within classified U.S. government systems in a matter of hours.
The Associated Press reported on June 23, referencing an unnamed official, that recent assessments undertaken in partnership with U.S. intelligence agencies revealed security flaws in sensitive and secure government computer networks.
Despite its impressive detection capabilities, the official noted that Mythos has not yet attained the sophistication necessary to exploit these security gaps within the same brief timeframe.
This evaluation was part of a broader initiative dubbed ‘Project Glasswing,’ a security endeavor in which Anthropic collaborates with reputable institutions and corporations.
Earlier, during a congressional hearing on June 11, Senator Mark Warner (D-Va.) highlighted these findings, stating, “This tool penetrated nearly all of our classified systems in just a few hours, not weeks.”
Senator Warner relayed this intelligence from General Joshua Rudd, leader of the National Security Agency (NSA) and U.S. Cyber Command, although the NSA opted not to comment on this revelation.
Anthropic has encountered friction with federal authorities, having been labeled a ‘supply chain risk’ entity after disputes arose with the U.S. Department of Defense regarding the parameters surrounding its AI applications.
In particular, following the launch of the Mythos 5 model and the Fable 5 model—which incorporates supplementary safety measures—the U.S. government imposed export control guidelines mandating that foreign nationals be barred from accessing these technologies due to national security apprehensions.
As a result, lacking the ability to verify user nationalities, Anthropic restricted access to both models for all users.
In a significant response, over 100 cybersecurity experts and notable figures from major technology firms, including Adobe and Nvidia, have penned a letter urging the U.S. government to rescind these control measures. They contend that such restrictions may inadvertently aid rival nations.
Source link: News.sbs.co.kr.
