Anthropic's Mythos AI model has uncovered significant vulnerabilities within classified US government systems during a recent testing exercise, raising alarms about cybersecurity preparedness.
Conducted under the auspices of Project Glasswing, the collaboration with US intelligence agencies aimed to proactively identify and mitigate risks before adversaries could exploit them.
Mythos broke into almost all of our classified systems, not in weeks, but in hours.
Senator Mark Warner, Virginia
This revelation, made by Senator Mark Warner during a congressional hearing, underscores the rapidity with which Mythos can detect vulnerabilities.
- Mythos identified vulnerabilities within hours, but it did not exploit them in that timeframe.
- The White House and Department of Defense have not commented on the findings.
The findings come at a time when Anthropic's relationship with the US government is under strain, following the company's refusal to allow its technologies to be used for domestic surveillance and autonomous weaponry.
In retaliation, the government placed Anthropic on a national security blacklist, severely limiting its operations, including a recent suspension of exports of its Mythos and Fable models.
This incident echoes broader concerns within the cybersecurity landscape, particularly as advanced AI models become integral to both defense and vulnerability assessments.