Anthropic says AI system identified thousands of critical software flaws
Cyber resilience concerns increase after Anthropic demonstrated large-scale AI vulnerability scanning capabilities.
Anthropic has published an update on Project Glasswing, a cybersecurity initiative focused on identifying software vulnerabilities using AI systems.
According to Anthropic, partner organisations used Claude Mythos Preview to identify thousands of high- and critical-severity vulnerabilities across software platforms and infrastructure systems.
The company said the initiative demonstrated how AI systems are increasing the speed and scale of vulnerability discovery processes. Anthropic reported that participating organisations observed substantial increases in software vulnerability detection capabilities during testing.
Evaluations cited by Anthropic suggested the system performed strongly in vulnerability identification and exploit-detection tasks compared with earlier AI cybersecurity models.
Anthropic also said the model analysed more than 1,000 open-source projects and identified vulnerabilities affecting widely used software components. The company highlighted a vulnerability identified in the open-source cryptography library wolfSSL as one example from the project.
According to Anthropic, the vulnerability was patched after disclosure.
Anthropic said AI-assisted vulnerability discovery may increasingly shift cybersecurity challenges toward verification, disclosure, and remediation processes. The company also said similar AI cybersecurity capabilities are likely to become more widely available across the industry.
Why does it matter?
The rapid growth of AI-driven cybersecurity is becoming increasingly important as AI is fundamentally changing the balance between cyber defence and cyber threats. Systems such as Anthropic’s Project Glasswing demonstrate that advanced AI models can identify software vulnerabilities at a speed far beyond traditional human-led security testing, potentially making critical infrastructure, financial systems, cloud platforms, and open-source software both safer and more exposed at the same time.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!