Apollo GraphOS Agent Services: Apollo Agrees Agents Need Explicit Rules — But Governance Stops at Access

On October 7, 2026, Apollo GraphQL unveiled GraphOS Agent Services at the Apollo Summit in San Francisco. This service is composed of search, identity, policy, and audit tools, which act between AI agents and enterprise systems. GraphOS coordinates over 2 trillion operations per month. Intuit is currently piloting the service in a preview phase.

Apollo claims that the rules regarding what is safe to return are typically based on a developer's judgment, not the API itself, and agents do not have access to this judgment. Therefore, these rules need to be made explicit and consistently enforced, regardless of the AI model's decisions.

Apollo's new governance system ensures that agents can only access what they are explicitly permitted to access. The system does not manage the financial aspect of agent transactions. According to a Gartner Market Guide for Guardian Agents from February 2026, 80% of unauthorized AI agent transactions in the next eight years will likely result from internal policy violations, such as oversharing, unacceptable use, or misguided AI behavior, rather than malicious attacks.

Apollo's access policy can determine whether an agent can see or do something, but it cannot decide whether an action should be paid for. The system includes an audit trail of all activities, which logs the instruction, authority, score, band, and outcome. The company has tested the system with two GraphOS-shaped instructions on the scriptmasterlabs.com/api/harness/decide endpoint, and the results show that the heuristic cannot differentiate between authorized but wrong actions and malicious behavior. In both cases, the system recommended a human review or holding the action for further inspection.