[DigitalToday reporter Chi-gyu Hwang (황치규)] Anthropic is expanding access to cybersecurity capabilities based on Claude Mythos5, SecurityWeek reported on Aug. 24 (local time).

The report said the move is based on the Glasswing project that began in April. It focuses on limiting direct access to Mythos-class models while increasing the outputs delivered to defenders.

The core of the Glasswing project is to provide Claude Mythos Preview and the follow-on model Mythos5 early to a small number of institutions. It aims to give defenders time to find and fix vulnerabilities before they spread widely or are exploited.

Anthropic described direct and unlimited access to the model as the most dangerous scenario. It said risk drops sharply if it provides only concrete defensive outputs such as patches or security alerts.

Anthropic is accelerating efforts to work with security partners to integrate Mythos5 into security operations, incident response and detection tools. End users do not interact with Mythos directly, and a dedicated interface runs the model in the background to provide only outputs such as a recommended patch list.

Claude Security, offered in beta to Claude Enterprise customers, has begun applying Mythos5 to code scanning. Scan results show CWE category, confidence, severity and remediation suggestions. Actual fixes are deployed only after going through Claude Code and being approved by a person.

Anthropic also introduced the open-source security fund called the Defender Advantage Fund (0xDAF). It will provide $35 million in Claude credits to organisations that help maintainers protect open-source projects.