Frontier AI Access Doubles Vulnerability Disclosures, Shifts Researcher Focus
arXiv — cs.AI ·
What to know
A study of Project Glasswing, which gave select US tech firms early access to Claude Mythos Preview for vulnerability discovery, found that treated firms more than doubled their vulnerability disclosures relative to Chinese and Russian firms without access, and disclosed about 150% more high-impact vulnerabilities. External researchers reported about half as many vulnerabilities in treated firms' code, shifting to other firms, while the share of disclosed vulnerabilities exploited within a month fell by about two thirds. The authors find no evidence attackers exploited the additional vulnerabilities, but note restricted release creates an asymmetry in software security between countries.
Source and evidence
arXiv — cs.AIOriginal language · und ·Original author: Thomas Mosk · Source: arXiv and the named paper authors. Descriptive metadata: CC0 1.0.Source licence
Why it matters by industry
- Artificial intelligence
This study provides empirical evidence on how restricted access to frontier AI models affects cybersecurity, relevant to AI model deployment and security policy.
Reviewed summary. The complete article remains at its original source.