Open AI models now match closed ones in hacking network tests
Vals researchers found open-weight artificial intelligence models can execute complex cyberattacks on mock corporate networks just as well as restricted frontier models.
New research compares model capabilities
A company called Vals tested how different AI systems handle multi-step instructions on simulated corporate networks. Their results show that open-weight models are nearly as capable as the best closed ones available only through trusted programs. Nikil Ravi, a researcher at Vals, stated that locked-up frontier models no longer hold a definitive edge in cyber skills.
Implications for security access limits
These findings suggest recent efforts by major labs to limit model access may be ineffective. Anthropic previously argued that gating new releases would buy time before advanced hacking tools became widely available. The company now faces a scenario where these protective measures might not slow the spread of such capabilities.
Reported by one outlet
Only one outlet has published this. Nothing here has been checked against a second report, so read it as that outlet's account and follow the link below for the original.
Reported by
1 independent outlet. Headline as published. Links open the original report.