Get the app

AI Has A Trust Problem. The Answer Is In The Chip

Forbes
Read full story· 5 min
Anthropic's Mythos AI model rapidly identified critical software vulnerabilities, highlighting how AI accelerates both discovery and exploitation of security flaws. This shift shortens the window for patching and raises concerns about AI agents' privileged access inside enterprise systems.

What happened

In April, Anthropic provided about 50 organizations that develop or maintain critical software with access to its new Mythos AI model as part of Project Glasswing. Within weeks, Mythos uncovered a decades-old vulnerability in OpenBSD and a bug in FFmpeg that had previously gone undetected by millions of automated tests. Shortly after, Google's Threat Intelligence Group identified the first known zero-day exploit created with AI assistance. These events demonstrated that AI can both rapidly identify and help exploit software vulnerabilities.

Key details

  • Mythos found a 27-year-old vulnerability in OpenBSD.
  • Mythos also detected a bug in FFmpeg missed by five million automated tests.
  • About 50 companies participated in Project Glasswing in April.
  • Google confirmed the first AI-assisted zero-day exploit one month after Mythos's release.
  • The median time to exploit a vulnerability dropped from 771 days in 2018 to near zero by 2025.

Why it matters

Security teams and software vendors now face much less time to patch vulnerabilities before they can be exploited, increasing the risk of attacks. Enterprises must also manage the new challenge of AI agents holding privileged access to sensitive systems and data, which can expand the potential impact of a breach. Organizations need to adopt new security measures that account for the speed and capabilities of AI, including stronger controls over agent access and technologies that protect data while it is being used.

#anthropic#ffmpeg#mythos#openbsd#software-security#vulnerability-management

More in Cybersecurity