Atlassian CEO Mike Cannon-Brookes says employees who direct workplace AI agents remain responsible for their output, including checking whether people receiving material drawn from internal documents have permission to see it. Fortune reported his comments Monday, quoting him as saying, “It has to be with the end user, ultimately.”
Atlassian began rolling out its Agentic Multiplayer Protocol (AMP) on Oct. 6, the company said. The framework is meant to bring agents into team workflows with identities and permissions that let organizations limit what each system can reach. Atlassian says AMP also records agent activity for teams to review.
The company’s announcement says an audit can identify the employee who triggered an agent and the data it touched. Atlassian discusses edit visibility separately: its design guidance calls for teams to see changes an agent makes, with contributions shown under the agent’s identity alongside the employee who directed it.
That guidance also covers access to apps and spaces. Atlassian’s announcement separately describes human checkpoints in agent workflows, placing review steps alongside activity records and permission limits.
The company says dedicated agent accounts and a non-human identity inventory are coming soon. Administrators are intended to use those controls to manage agents’ access and revoke it when needed.
