AI safety company Anthropic has revealed that its advanced AI cybersecurity initiative, Project Glasswing, has already identified more than 10,000 high- and critical-severity vulnerabilities across widely used open-source software projects. The discovery highlights how rapidly artificial intelligence is transforming cybersecurity and software protection.
Project Glasswing Is Focused On Securing Critical Software
Anthropic launched Project Glasswing as a restricted-access cybersecurity initiative involving its powerful AI model called Mythos Preview. The project gives select organizations access to the AI system to help detect vulnerabilities in critical software infrastructure before hackers can exploit them.
According to Anthropic, the AI model has been deployed across more than 50 partner organizations and software ecosystems. The vulnerabilities discovered include flaws in operating systems, developer tools, open-source libraries, cloud infrastructure software, and internet-facing systems.
AI Is Finding Bugs Faster Than Humans
One of the biggest revelations from the project is the sheer speed at which AI can discover vulnerabilities compared to traditional security teams. Industry experts say advanced AI systems can now analyze massive codebases, identify unusual patterns, trace data flows, and detect hidden weaknesses far faster than human researchers.
Reports suggest the Mythos model even discovered decades-old software bugs that had previously escaped detection despite years of audits and testing.
Cybersecurity analysts say this could fundamentally change how software security works in the future, where AI continuously scans and protects digital infrastructure in real time.
Open-Source Software Faces New Pressure
The findings also expose the growing security challenges surrounding open-source software, which powers much of the modern internet, enterprise systems, cloud platforms, and AI infrastructure.
Many open-source projects are maintained by small volunteer teams with limited resources, making it difficult to patch vulnerabilities quickly even after they are identified.
Security experts warn that the traditional “patch window” — the time between discovering and fixing vulnerabilities — is shrinking rapidly because AI systems can now uncover flaws at unprecedented scale.
AI Cybersecurity Race Intensifies
The development also highlights the growing AI arms race in cybersecurity. While companies are using AI to strengthen defenses, experts fear malicious actors could also deploy advanced AI systems to discover and exploit vulnerabilities faster than ever before.
Anthropic says Project Glasswing is designed with strict safeguards and limited access to prevent misuse of its powerful vulnerability-detection capabilities.
The company believes AI-assisted security systems could become essential for protecting global digital infrastructure as software complexity continues to increase.
AI Coding And Security Tools Becoming Mainstream
AI-powered coding and security agents are rapidly becoming one of the hottest sectors in the technology industry. Companies including OpenAI, Google, Microsoft, and Anthropic are heavily investing in AI systems capable of writing code, auditing software, and detecting vulnerabilities automatically.
Experts believe future software development may increasingly involve AI agents continuously monitoring, testing, and securing codebases alongside human developers.