Rapid News Brief

Zscaler said Nov. 20 that it is working with Anthropic and Mistral on enterprise AI security. Its announcement separates access for the company’s own vulnerability discovery and research from a model integration intended for customer security tools. Zscaler scheduled further AI-security announcements for Dec. 9 at 1 p.m. Eastern time.

Access and verification

Zscaler said Claude Mythos 5.1 was integrated into Endpoint AI Security to power Exploit Paths, which correlates evidence across attack sequences. A limited preview is planned. The announcement also describes a Mistral integration through Zscaler’s sovereign Zero Trust Exchange in Europe. It describes application-specific access rather than network-wide access.

Anthropic’s Cyber Verification Program provides context for the research-access side. In an Oct. 6 announcement, updated Nov. 18, Anthropic describes tiers with different verification requirements and security controls. Defense Access covers activities including incident response and vulnerability validation. Red Team Access adds penetration testing, but only against systems organizations are authorized to test.

Anthropic says applicants must provide proof of the controls required for their tier, and administrators must assign access to specific workspaces. The company also says operating critical infrastructure alone does not qualify an organization for its highest-access tier; applicants must establish authorization and explain the work requiring that access. Anthropic says real-time blocks still apply to actions that could cause physical harm or mass disruption.

Region and model choice

Mistral’s separate Aug. 11 infrastructure announcement describes regional inference endpoints in Europe and the United States as generally available through its hosted platform. It says inference and associated processing occur in the selected region, subject to limited, safeguarded transfers to subprocessors outside that region.

That account concerns Mistral’s own platform, not independent confirmation of how Zscaler’s integration handles each customer’s data. Mistral also describes model choice separately from processing location, with plans for third-party open models on the same infrastructure and regional controls.