Senate testimony alleges AI model breached NSA systems in hours, prompting US restrictions

The US government ordered Anthropic to restrict access to its Fable 5 and Mythos 5 AI models, limiting them to US citizens only. This decision followed the discovery of a jailbreak, which, according to Senate testimony, allowed Anthropic's Mythos model to penetrate nearly all classified systems managed by the NSA and US Cyber Command within hours, with further coverage provided by Security Affairs.

Senate testimony claims Anthropic's Mythos AI model breached NSA and Cyber Command systems in mere hours, leading to a US-ordered shutdown of the model. The US government directed Anthropic to restrict access to its two most capable models, Fable 5 and Mythos 5, exclusively to US citizens. Due to the impracticality of real-time nationality verification, Anthropic shut down both models for all users, including allies. Senator Mark Warner cited General Joshua Rudd, who leads both the NSA and Cyber Command, stating that Mythos infiltrated classified systems rapidly. While these claims are unverified, Mythos's capabilities are not in dispute; it has reportedly uncovered thousands of vulnerabilities, including a 27-year-old flaw in OpenBSD.

The shutdown impacts cybersecurity research, potentially hindering defenders while limiting the risk of malicious actors exploiting advanced AI. This event highlights the growing geopolitical significance of advanced AI capabilities and raises concerns about data and analytical sovereignty for nations like those in Europe, which regulate AI but lack comparable frontier models.

Source: Security Affairs