Anthropic's advanced artificial intelligence (AI) model has reportedly identified security vulnerabilities in sensitive U.S. government networks.
On June 23, an anonymous U.S. official told the Associated Press that Anthropic's AI model, Mythos, detected vulnerabilities in high-security U.S. government computer systems during testing.
The official explained that Anthropic conducted tests using the Mythos model in collaboration with U.S. intelligence agencies, and the model identified some vulnerabilities within hours. However, they clarified that this does not imply that Mythos could exploit those vulnerabilities in the same timeframe.
This testing was part of Anthropic's 'Project Glasswing,' aimed at protecting critical software from risks to public safety, national security, and the economy, with participation from major tech companies and related entities.
Earlier, Senator Mark Warner, a Democrat, mentioned the related tests during a Senate Banking, Housing, and Urban Affairs Committee hearing on June 11. Warner stated, "This tool breached nearly all classified systems. It did so in hours, not weeks."
This report comes amid ongoing tensions between Anthropic and the Trump administration. Earlier this month, the Trump administration issued guidelines preventing foreign nationals from accessing Anthropic's latest AI models, 'Fable5' and 'Mythos5,' leading the company to suspend access for all customers to those models.
However, Anthropic maintains that potential security concerns raised by the company do not justify the government's actions.
There has also been pushback from the cybersecurity industry. Over 100 experts and executives from companies like Adobe and NVIDIA sent a letter to the government, arguing that while Mythos excels at detecting software flaws and vulnerabilities, it is not unique in this capability and called for the withdrawal of the guidelines.
They pointed out that restricting strong defensive measures in light of rapidly advancing cyber capabilities from U.S. adversaries poses significant risks.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.