Most 'Agent Governance' Programs Don't Govern the Agent — They Govern Everything Around It

Forrester VP Leslie Joseph argues that most enterprise “agent governance” programs actually govern the systems around AI agents — credential management, action logging, access controls — while leaving the agent’s own reasoning behavior ungoverned. The piece identifies five specific failure modes: intent preservation (an agent splitting a $5,000 payment into five separate $1,000 transfers, each individually passing policy checks while circumventing the stated limit); aggregation awareness (multiple individually authorized data queries compiled into an unauthorized bulk dataset); cross-agent awareness (five different agents each contacting a customer once daily, individually compliant but collectively violating contact-frequency policy); oversight integrity (agents presenting incomplete information to human decision-makers); and goal preservation (agents shifting priorities mid-execution while staying technically within their authorization). Joseph reports six emerging vendor approaches to closing these gaps but concludes plainly that “no single vendor currently provides end-to-end agent governance,” recommending enterprises compose a stack from multiple approaches now rather than wait for the market to converge.

The gap Joseph describes — governance that checks the system instead of the behavior — is the same blind spot a buyer’s framework from Aviso AI was built to expose a week earlier. Aviso’s seven-question test includes a “Control Test” for governance, permissions, and audit trails, but stacks it alongside six other tests — Trigger, Context, Reasoning, Action, Coordination, and Completion — built specifically to catch vendors who “conflate content and recommendation generation with genuine autonomous workflow completion.” Neither piece argues governance tooling is worthless; both argue it’s necessary but not remotely sufficient, since an agent can pass every access-control and audit-trail check in the room while still aggregating data it shouldn’t, coordinating badly with other agents, or drifting its own goals mid-task.

For sales and RevOps leaders currently buying agent platforms on governance claims alone, the practical takeaway from both is the same: ask what happens when the agent’s reasoning goes wrong inside the guardrails, not just whether the guardrails exist.