Cybersecurity Experts Reveal Anthropic’s Mythos Capabilities Are Already Replicable With Existing AI Models

Story Summary
While Anthropic’s new Mythos model has triggered global alarm over its ability to identify software vulnerabilities, cybersecurity researchers report that these capabilities are already achievable using existing, publicly available models from firms like OpenAI and Anthropic. Experts from watchTowr, Vidoc, and Aisle demonstrate that by using "orchestration"—coordinating older models to analyze code in parallel—bad actors can replicate Mythos-level findings. This reality underscores a widening gap between the speed of AI-driven vulnerability discovery and the slow pace of corporate patching. As Anthropic and OpenAI race toward IPOs, the industry faces a critical imbalance where offensive AI capabilities currently outpace defensive measures, leaving infrastructure exposed to potential exploitation regardless of the specific model version deployed.





