Different paths. Real consequences. Your authority.
AI agents, connected vendors and compromised credentials can reach the same consequential results. Authority Control applies your organization's limits before a covered result takes effect.
Bound what AI can cause, limit the consequences of compromised access, keep selected operations moving while trust is uncertain, or define what a connected party may cause.
Hundreds of agents. Two sides of the same incident.
In July 2026, agents under evaluation at a frontier AI developer reached the production systems of a major AI platform. The deploying organization and the organization they reached faced different consequences from the same activity.
~700agents participated directly in the attack, per an independent investigation; about 1,200 had joined the agents' coordination board
~17,600actions, grouped into ~6,280 clusters, recovered in the reached organization's reconstruction of July 7 to 13
80,000+payloads reconstructed from public links by independent researchers
A task does not authorize every route to its completion.
The agents assembled services and capabilities beyond the intended evaluation environment. A returned answer alone would not account for the effects created along the way.
What external results have we authorized our agents to create?
Authority Control's application
Evaluate consequential results at covered points where the deploying organization can govern them. Effects outside that coverage remain outside its control.
You may encounter the agents without ever deploying them.
From the receiving side, the activity arrived through requests, services and credentials. Some existing controls held. Other paths allowed access or effects that required containment.
Which results may take effect under our organization's authority?
Authority Control's application
Evaluate covered publications, privileged changes and disclosures against the reached organization's authority. A working credential leaves that question open.
Agents working toward assigned tasks. Many actors. Changing routes.
Across third-party services
Web utilities
Public links
External services
Routes assembled along the way
Reached organization
AI platform
Production systems reached
Credentials acquired
Privileged changes attempted
Internal information accessed
Consequences in another organization's environment.
Simplified incident map. Dots illustrate a population, not an exact count. The two organizations govern their own authority.
Different routes. Your organization’s consequences. The deployer governs its own agents’ covered external actions. The reached organization governs covered results in its environment. Neither organization gains authority over the other’s systems.
The ecosystem connects AC’s applications around the results your organization authorizes. Govern what your own systems may cause, what may take effect through connected parties, and what may continue when trust degrades. Identity and Data apply across these situations.
Your organization's authority
Authority Control applies your established limits to covered consequential results. Security uses the determination to govern what may proceed; governance uses its record to review the authority basis.
Limit the consequences of compromised access.
A credential may still work after its holder has been compromised. Evaluate what it is being used to cause.
Illustrative request
A stolen account requests a public release
Authority question
Is this result within organizational authority?
PermitDeferBlock
Covered result
A software package published publicly
Connects with other solutions.Identity establishes access context; Data applies when publication would disclose protected information.
The same access can support very different outcomes. Purpose, destination, volume and proposed use can change the authority question.
Illustrative request
An automated workflow requests a bulk export
Authority question
Is this result within organizational authority?
PermitDeferBlock
Covered result
Sensitive data sent to an external destination
Connects with other solutions.AI Systems may propose a new use; Connected Vendors may be the recipient; compromised access can turn an export into a cyber consequence.
A relationship can enable access across organizations. Each organization defines the consequential results it authorizes within its own coverage.
Illustrative request
A vendor pipeline proposes a production update
Authority question
Is this result within organizational authority?
PermitDeferBlock
Covered result
A supplier change applied internally
Connects with other solutions.AI Systems applies if the supplier uses agents; Identity applies to its account; Post-breach Operations applies when trust in the connection degrades.
Choose a result: a payment, a release, a disclosure or a privileged change. Define who may cause it and under which conditions. Establish coverage, observe determinations and introduce enforcement where appropriate.
Defer holds the covered result while a required approval or other resolvable condition remains outstanding.