Payward Leverages AI Model for Enhanced Cybersecurity

Claude Mythos 5 is considered one of the strongest AI models in Anthropic's arsenal, capable of inspecting code, identifying weaknesses, suggesting patches, and assisting researchers with testing exploit paths. Project Glasswing restricts access to vetted organizations, which must meet security requirements before receiving access to the model.

Payward plans to use Mythos 5 to scan its software environments for vulnerabilities over the coming weeks. Findings will enter Payward's existing security review process rather than automatically producing software changes. The company intends to disclose validated vulnerabilities affecting third-party open-source projects to their maintainers, aiming to protect other organizations using the same software.

The announcement does not specify whether Mythos 5 will receive access to production systems, isolated copies of source code, or controlled testing environments. Payward's co-chief executive Arjun Sethi stated that AI can change the balance between attackers and defenders by reading code at a greater scale, but human review is still required before classifying an issue as a vulnerability.

Anthropic has limited Mythos 5 access due to its dual-use capabilities, which can also help produce working exploits. Participating organizations must accept 30-day data retention for safety monitoring, and the company requires customers to acknowledge this condition.