A Microsoft Copilot Studio agent that goes wrong can now be restored to a known-good state, including its configuration, logic, prompts, topics and flows. The capability is part of the AvePoint Trust Platform, which AvePoint has launched to govern AI agents across vendors.
The platform discovers sanctioned and shadow agents, assigns each an owner and a policy, and produces audit-ready evidence. AvePoint says it operates independently of any single cloud, model or AI provider.
The launch is aimed at the data exposure that agents create. A copilot or agent reads everything it is permitted to read, and can act on it thousands of times an hour without breaking a rule.
AvePoint's State of AI 2026 Report found 88.4 per cent of organisations had at least one AI agent-related security incident in the past 12 months. That included organisations confident they could prevent unauthorised AI access to data.
The platform is built around four disciplines. Observe discovers agents, including those built by citizen developers, and shows which content is overshared, stale or unlabelled before an agent finds it.
Protect provides backup and recovery for Microsoft 365, Power Platform, SaaS applications such as ServiceNow, and Azure, AWS and Google Cloud. It now extends to the agents themselves.
Govern gives each agent an owner, a policy and data sensitivity context. Business owners attest to and recertify agents through Slack, Teams and email, from discovery to retirement.
Optimise removes unnecessary content, improves prompt effectiveness and tracks token consumption and budgets.
The four sit on what AvePoint calls a Governance Graph. It joins regulatory frameworks such as the EU AI Act, ISO 42001 and the NIST AI Risk Management Framework with business context. That context includes owners, policies, information barriers and the purpose of each workspace.
New governance and security capabilities include an AI Inventory, an AI Dependency Graph and Shadow AI Detection. Together they discover agents, models, tools and MCP servers, map the relationships among them, and identify unsanctioned activity.
Enhanced policies flag ownerless agents and monitor agent access to sensitive content. Expanded data security posture management and Kinetic Classification can identify and tag AI-generated content and connect classification to automated remediation.
The platform can also issue signed charters that define what each agent may do, and roll back changes an agent has made.
Discovery now covers agents built with Anthropic's Claude and connected MCP servers, alongside Copilot Studio, Microsoft Foundry, Google Vertex AI and Salesforce Agentforce. Windows and server-hosted agents are also covered.
A separate AI Cost Optimisation module brings spend, usage and adoption data together for IT and finance teams. It replaces AI budget requests handled by email and spreadsheet with a structured approval process, and nudges staff where licensed tools go underused.
"Every organisation now has access to the same AI capabilities. What truly sets them apart is how far that AI can safely reach, what permissions it actually has, and whether there's a way to reverse course when it makes a wrong call," said Dr Tianyi Jiang, CEO and co-founder of AvePoint.
The platform is in private preview, with sign-ups open for early access.