The Mythos model created by Anthropic was able to breach 'almost all' of the US National Security Agency's classified systems in a matter of hours rather than weeks, according to Senator Mark Warner (D-VA), vice chairperson of the Senate Intelligence Committee, as reported by the Associated Press. During the June 11 briefing, General Joshua Rudd, the chief of the NSA and US Cyber Command, made the claim, which The Economist later published. The compromise's scope is still unknown, with neither the specific systems impacted nor the exact techniques Mythos used made public. The depth of Mythos' powers has been made clear by this recent discovery, demonstrating the AI model's extraordinary capabilities in gaining access to US government classified networks.
Anthropic teamed up with Washington's intelligence agencies to conduct tests using Mythos under Project Glasswing, a restricted program designed to find and fix vulnerabilities in critical software before attackers could exploit them, according to the Associated Press. The testing exercise was conducted under controlled conditions to identify potential security weaknesses in government systems. Although Mythos identified certain vulnerabilities within hours, that did not mean the model was able to exploit them within that time, citing an unidentified US official. The official explained that Anthropic opted against a public release for Mythos and instead granted select access to more than 50 technology firms, allowing them to identify and remediate critical software vulnerabilities. The controlled nature of this testing demonstrates how AI models are being used proactively to identify and address security vulnerabilities before they can be exploited by malicious actors.
Since the Anthropic agreement with the Pentagon went awry, the US has been extremely busy with recent developments including the prohibition on exporting Mythos-class models to foreign nations. Due to the inability of the AI company to confirm a person's nationality before using its most sophisticated models, Anthropic ultimately decided to discontinue Claude Fable 5, a restricted Mythos model. The company has vigorously opposed a long-term recall of its Mythos-class models, contending that shutting down its most sophisticated capabilities due to insignificant compliance concerns will hinder Western AI progress and make vital infrastructure more susceptible to foreign competitors who do not encounter similar legal obstacles.
The US government's decision to restrict foreign access to Anthropic's most advanced AI models has created significant international tensions. US Commerce Secretary Howard Lutnick disclosed that the action was taken because the officials feature the AI systems could be used by military or intelligence agencies in Russia, China and other countries considered security risks, according to Reuters. The restrictions have affected governments, financial institutions, research organisations outside the United States, including some close allies. Even members of the Five Eyes intelligence alliance, including Britain, Canada, Australia and New Zealand, were reportedly impacted by the decision. Several tech leaders have criticised the move by the US government to ban these powerful AI models for non-American users, hinting that AI power should not be concentrated within a limited group of people.
