Delinea has joined Anthropic’s Project Glasswing, an initiative designed to strengthen critical software using frontier AI. The identity security company will test Claude Mythos 5.1 against its own security-critical code to identify vulnerabilities before attackers can exploit them.
The collaboration focuses initially on Delinea’s software for vaulting, storing, brokering and rotating privileged credentials. Identity security is particularly sensitive because weaknesses in privileged-access systems can potentially expose multiple systems they protect.
Delinea will apply the AI model to credential and secrets handling, cryptographic implementations, session brokering, privileged-access paths and authorization enforcement. The objective is to complement existing security reviews and adversarial testing with AI-powered vulnerability discovery.
Importantly, Delinea says Mythos will operate under privileged-access principles itself. Its access will be scoped, temporary and logged, with testing conducted in an isolated environment where customer data is excluded.
Validated findings will enter Delinea’s existing product-security process for remediation. Recurring vulnerabilities can subsequently influence secure coding standards, pre-merge checks and testing, strengthening future software releases.
The initiative signals an important cybersecurity shift: frontier AI is becoming both an offensive risk and a defensive weapon. Security vendors can increasingly use powerful AI models to attack their own software first—turning machine-speed vulnerability discovery into proactive cyber defence.
