The US government’s ban on Anthropic’s flagship AI model, Mythos 5, has raised concerns about its security and capabilities. According to a report by The Economist, the model was able to break into “almost all” classified systems of the National Security Agency (NSA) within hours during a controlled red-team test. The breach occurred when the model was paired with other defensive tools under specific simulated environmental conditions. Anthropic has claimed that the breach was a narrow jailbreak and not a genuine autonomous offensive intrusion.
The ban on foreign nationals accessing Fable 5 and Mythos 5 models has been imposed due to national security concerns, but Anthropic argues that it cannot practically enforce nationality-based access restrictions without pulling the systems for everyone. The company is working with the White House to restore access and develop a collaborative risk-management framework.
Public reaction to the news has been divided, with some criticizing the government’s cybersecurity and others questioning the validity of the claim. Cybersecurity experts have pointed out that AI capabilities have compressed attack timelines from hours to minutes, making it a growing concern.
Source: https://www.tomshardware.com/tech-industry/artificial-intelligence/anthropics-powerful-mythos-ai-reportedly-breached-almost-all-nsa-classified-systems-within-a-few-hours-during-red-team-test-report-sheds-more-light-on-the-u-s-governments-sudden-ban-on-the-flagship-models