Anthropic's Mythos AI exposes flaws in classified US systems
What's the story
Anthropic's artificial intelligence (AI) model, Mythos, has exposed vulnerabilities in highly sensitive US government computer systems. The revelation was made by a US official who spoke on the condition of anonymity to the Associated Press. The official said that the testing was part of a collaboration between Anthropic and US intelligence agencies to identify potential flaws using the new cybersecurity-focused Mythos model.
Rapid detection
Mythos flagged vulnerabilities but didn't necessarily exploit them
The official clarified that while the Mythos model flagged certain vulnerabilities within hours, it doesn't necessarily mean that it exploited them in the same timeframe. The testing was part of an Anthropic initiative called Project Glasswing. This project brought together tech giants and other companies to secure critical software from potential risks posed by the Mythos model to public safety, national security, and economy.
Company stance
Anthropic has disabled Fable and Mythos models
In light of the Trump administration's directive, Anthropic has disabled its Fable 5 and Mythos 5 models for all customers. The move was made to comply with a directive from the Trump administration that restricts foreign nationals from using these advanced AI models. However, Anthropic disagrees with the government's decision, arguing that it was not warranted by the potential security issue it flagged.
Expert opinion
Cybersecurity experts urge Trump to lift directive
A group of over 100 cybersecurity experts and leaders from companies like Adobe and NVIDIA have urged the Trump administration to lift its directive. They argue that the move could help US adversaries more than it hurts them. The letter said while Anthropic's Mythos models are "quite good" at finding flaws in software and weaponizing exploits, they are "not uniquely good at these tasks."