Concerns Over Anthropic’s “Mythos” AI Model
Anthropic’s advanced “Mythos” AI model has reportedly uncovered vulnerabilities in highly secure U.S. government systems in a matter of hours, raising significant security alarms.
In tests carried out by U.S. intelligence agencies, Mythos pinpointed vulnerabilities in government infrastructure rather quickly. However, it’s important to note that this speed doesn’t necessarily mean bots can exploit the vulnerabilities at a similar pace, as mentioned in a recent report.
This testing aligns with Anthropic’s Project Glasswing, which aims to collaborate with various organizations and institutions to tackle possible security issues. There are even claims suggesting that both Mythos and another model, “Fable,” could lead to catastrophic outcomes in AI development.
Some observers express skepticism over these dire warnings, suggesting they might primarily serve to market Anthropic amid stiff competition in the AI landscape.
However, during a Congressional hearing earlier this month, Sen. Mark Warner, who leads the Senate Intelligence Committee, revealed that Gen. Joshua Rudd, director of the National Security Agency, indicated Mythos had “compromised nearly all of our nation’s classified systems in hours, not weeks.”
Meanwhile, White House officials informed that various U.S. government sectors are leveraging state-of-the-art AI models to locate and counteract cyber vulnerabilities, asserting that measures are being taken to secure sensitive systems.
Both the National Security Agency and Anthropic have yet to respond to requests for comments regarding the situation.
Earlier this month, the Trump administration introduced export controls on “myths” and “fables” following warnings from Amazon CEO Andy Jassy about potential weaknesses in security barriers.
In response to these concerns, Anthropic dispatched several executives to Washington, D.C., in an effort to gain governmental support and took its models offline, claiming that this was the only way to comply with the newly imposed restrictions.
During the model’s shutdown, the National Security Agency was testing Anthropic’s latest AI tool, highlighting the essential role that such bots are already playing in cybersecurity and national security initiatives, according to reports.
Relations between Anthropic and the U.S. government have been strained as Anthropic CEO Dario Amodei declined to provide the Pentagon unrestricted access to AI tools during contract discussions, seeking to impose limits on mass surveillance and weaponization.
After the government enacted export restrictions on Anthropic’s newest models, the company asserted that it was simply signaling potential security risks associated with bots and labeled the measures as unnecessary.
Despite this, White House officials expressed anger at Anthropic’s assertion that these risks were a “narrow” issue, despite long-standing warnings about the serious implications of unchecked AI bots.
Now, Anthropic appears to be seeking to mend its relationship with the government, recently offering to collaborate more closely with the White House. Reports suggest that discussions between Anthropic and Trump officials are advancing positively, though a definitive timeline for resolving these issues is still uncertain.
At the recent G7 summit in France, President Trump acknowledged that dialogue with Anthropic was “progressing well,” while Prime Minister Amodei urged world leaders not to succumb to protective instincts when it comes to AI regulations.
