Netskope and Glasswing detail Claude Mythos Preview vulnerability work
Netskope’s blog describes its collaboration with Anthropic’s Project Glasswing to access Claude Mythos Preview for vulnerability identification, while arguing that AI-enabled discovery will also increase attacker capability across enterprise environments.
Research Overview
The post frames Mythos Preview as a development that affects how vulnerabilities may be found in code at scale, tying the update to broader frontier model releases.
It also positions the enterprise risk question as focused on where exposure occurs when AI capabilities become widely available to different actors.
Key Findings
Netskope reports observations from Netskope Threat Labs, stating that, among tracked enterprises, the number of AI applications in use grew fivefold over the past year.
The company also says the average organization tripled its AI user base to 37 deployed AI agents and saw 223 AI data policy violations per month.
Operational Impact
The blog emphasizes a defense-in-depth approach based on real-time governance and security that accounts for AI and cloud workflows, including zero-trust treatment for every agent.
It states that each transaction, including those originating from people, devices, or AI agents, should be inspected, governed, and controlled in real time, describing the challenge as involving data and identity alongside transactions and code.
Product Update and Collaboration
Netskope says it is using the Project Glasswing collaboration to access Claude Mythos Preview to identify and fix vulnerabilities in Netskope software, and it plans to share findings with the Glasswing coalition and the security community.
It also notes that earlier in the month it launched AgentSkope for deploying AI agents across security and networking workflows and announced an integration with Anthropic’s Claude Compliance API to connect Netskope data governance and compliance controls with Claude Enterprise usage. The post quotes Michael Moore, Anthropic’s Head of Cybersecurity: “Security teams need AI that can triage noise, connect signals across systems, and explain its reasoning to the analyst on the other end. Claude is built for that kind of work, and Netskope brings the platform, data, and SecOps expertise to apply it across security workflows. AgentSkope is a strong example of how the two can help teams keep pace with today’s threats.”
This blog centers on Netskope’s Project Glasswing work with Claude Mythos Preview and links it to real-time, zero-trust governance for AI-heavy enterprise traffic, while citing Netskope Threat Labs metrics on AI application growth and policy violations. Blog Signals brief is a fact-based summary of the vendor blog.