Summary
- AvePoint announced its Trust Platform on 27 October, with access offered through private-preview registration.
- It says the system inventories AI agents, models and MCP servers across several vendor ecosystems.
- AvePoint’s own 2026 survey reports 88.4% of respondent organisations experienced an agent-related security incident; methodology is not independently verified here.
AvePoint has announced a platform designed to give organisations a consolidated view of AI agents, the data those agents can reach and the policies governing their actions across different technology suppliers. The AvePoint Trust Platform was introduced on 27 October with new capabilities spanning AI inventory, security, governance, recovery and spending controls.
The company says the product can discover agents, models, tools and Model Context Protocol servers and map their relationships to information repositories and business owners. Its announced coverage includes agents built with Microsoft Copilot Studio and Foundry, Anthropic Claude, Google Vertex AI and Salesforce Agentforce, alongside Windows and server-hosted environments.
AvePoint invited organisations to register for a private preview and said it would demonstrate the platform at Microsoft Ignite in November. Neither the announcement nor the demonstrations establish general availability. Discovery, policy enforcement and restoration capabilities across all named ecosystems remain vendor descriptions pending independent deployment evidence.
Mapping access across agent ecosystems
An enterprise AI assistant may use files, messages, software tools and structured business data supplied by different systems. The permissions associated with those connections determine what information the assistant can retrieve and what operations it can attempt. If an agent inherits overly broad access from a user or service account, the immediate concern may be authorised but inappropriate use rather than an external hacker crossing an access barrier.
AvePoint’s proposed AI Inventory and Dependency Graph seek to record those relationships. The company also describes shadow-agent detection intended to identify automated tools created outside central procurement and security processes. Such a register could support governance only if discovery is sufficiently complete and the records stay aligned with changing permissions and deployments.
The platform groups its functions under observing, protecting, governing and optimising. The governance portion includes assigning an owner, policy and data sensitivity context to each agent, with recertification activities delivered through common collaboration tools. The security component includes classification and monitoring of sensitive content.
Recovery and accountability
AvePoint also highlights the ability to recover agent configurations and other components after an unwanted change. It cites Microsoft Copilot Studio agents as an example where configuration, prompts, workflows and associated logic can be restored to an earlier state. That is distinct from proving the actions already carried out by an agent can always be reversed.
An AI agent may have sent an email, changed an external record or passed information to a third-party service before its own configuration is restored. Recovery of the software object is therefore only one part of operational resilience. A complete incident response may still require tracing external effects and dealing with information that cannot be recalled.
The company uses the term AI Trust to describe governed data access, least privilege, policy enforcement, audit evidence and recoverability. Its supporting Governance Graph is presented as a way to connect regulations and standards, including the EU AI Act and ISO 42001, to an organisation’s own roles and information barriers. Mapping policy to a system is useful for documentation, but regulatory compliance depends on how controls operate in practice and the duties of the organisation deploying them.
AvePoint also announced AI cost optimisation functions that connect usage and budget requests to approval workflows. Spending visibility is part of the commercial proposition rather than evidence, by itself, that security outcomes have improved. The financial and security components should be assessed against separate performance measures.
What remains unproven
AvePoint cites findings from its State of AI 2026 survey, commissioned by AvePoint and conducted with Osterman Research, originally published in June. Based on responses from 750 enterprise leaders across the Americas, EMEA and Asia-Pacific, it reports that 88.4% of respondents’ organisations experienced at least one AI agent-related security incident during the preceding 12 months. This is a respondent-reported figure from a commissioned survey, not independent verification of breaches or a representative prevalence estimate across every enterprise.
The offering spans multiple agent ecosystems rather than operating only within a single cloud provider’s product. AvePoint says that mapping the links among agents, tools, sensitive data and owners is part of its governance design, but the breadth and accuracy of that mapping have not been independently demonstrated.
The platform’s private-preview phase leaves open questions about discovery accuracy, policy enforcement at the moment of an action, coverage of agent-to-agent connections and the extent to which its audit trails can reconstruct real-world changes. Those are product evaluation questions, not confirmed weaknesses in the system.
The 27 October launch establishes a private-preview offering spanning AI estate discovery, access governance and recovery. How completely it discovers unsanctioned agents, enforces permissions at the moment of action and records downstream effects remains to be tested in customer environments.
