Enterprise AI policies are already falling behind agents

Many enterprise AI governance policies were written for a simpler problem: stopping employees from pasting sensitive company data into public models. According to the author, Chief People & AI Transformation Officer at Zapier, that approach is now badly out of date as companies deploy AI agents that can query databases, update records, and trigger workflows across connected systems.

The core problem is that older frameworks were built to limit exposure, not manage autonomous work. That leaves companies with policies too vague to control what agents can actually do at the system level. In the author’s view, governance only works if it translates into concrete constraints, including which systems an agent can access, what actions it can take, and under what conditions.

The piece lays out four questions companies should ask when auditing their governance:

• Can employees quickly see what AI tools can access on their behalf?
• If an agent makes a bad decision, how fast can the company revoke its access?
• Does policy define approved uses, not just banned ones?
• Does governance specify agent permissions at the system level?

To address the first issue, the author recommends maintaining a permissions inventory covering approved AI tools, their connected systems, authorized actions, and the team or individual responsible for each integration. That record can be managed manually or through an AI governance platform, but it needs to stay current and easy to find.

For revoking access, the argument is for centralized authentication rather than credentials scattered across scripts and sessions. The article points to the Model Context Protocol (MCP) as a standard designed to give agents a structured, auditable way to access external systems through OAuth instead of embedding credentials in prompts or scripts.

Recommended reading

Rosatom to build CNC motors in Russia by 2027

The author also argues that governance should define what is permitted, not only what is prohibited. For agents, broad restrictions are not enough; they need explicit boundaries around approved tools, system connections, and authorized actions. In practice, that means provisioning access through identity and access management systems, assigning each agent a defined role with scoped permissions, and logging every action it takes.

The article’s bottom line is that governance has to be treated as an ongoing operational process. As agent capabilities change, companies need to review access definitions, audit permissions, and update permitted-use frameworks fast enough to keep up.