Anthropic acknowledged that human error in configuring its content management system exposed details of an unreleased AI model and other internal materials through an unsecured, publicly searchable data trove. The company said it was testing a new model described externally as a “step change” in capability and indicated the trial involved early access customers. Security researchers reviewed roughly 3,000 related assets found in the cache, including draft blog content reportedly naming the model and internal images and documents tied to an invite-only CEO summit. After being notified, Anthropic removed public search access to the cache. The incident highlights a tangible cybersecurity and governance risk for higher education partners evaluating AI vendors, especially where institutions buy custom tools and need assurances about data handling, access controls, and operational security. For campuses, the immediate implication is procurement due diligence: vendor incident response and the ability to prevent inadvertent disclosure of model and internal program information.