Human oversight, as a thing you can actually ship.

High-risk systems must be designed and developed so that natural persons can effectively oversee them during use. That includes understanding the system's capacity and limits, staying alert to automation bias, correctly interpreting output, and being able to decide not to use the output or to intervene and interrupt.

Interrupting becomes a call your agent makes and waits on. Pushary pauses the run, sends one short, specific question to a named person on the device they carry, and returns their choice. Override is one of the answers they can pick, just like Accept.

High-risk systems must technically allow for the automatic recording of events over the lifetime of the system, to a degree appropriate to the intended purpose.

Pushary saves each decision with the question as the person saw it, the option they chose, who they are, and the time. Pushary keeps it for your plan's history window (90 days on Partner Launch, 365 on Partner Growth), and you can query and export it. Add a callback URL and each answer is also sent to your own server as a signed webhook, so your own evidence store keeps a copy.

Deployers of high-risk systems must assign human oversight to natural persons who have the necessary competence, training and authority, and must make sure those people have the support to carry it out.

Each question goes to one specific person, by the id you already use for them, rather than to a shared channel. Two approvers on the same action means two separate calls, and two saved decisions, each tied to its own person.

Providers must make sure people are told when they are interacting with an AI system, unless that is obvious. They must also make sure synthetic audio, image, video and text output is marked in a machine-readable way. Deployers must disclose deep fakes. They must also disclose AI-generated text published to inform the public on matters of public interest, unless a person holds editorial responsibility for it. These duties took effect in August 2026 and sit outside the high-risk regime.

Pushary does not solve this one, and we would rather say so plainly. Marking your own model output belongs to whatever generates it. Pushary sits at the approval step, which is a different obligation. The difference matters, because the transparency duties apply now while much of the high-risk timetable does not.

The annex lists the use cases treated as high-risk. They include areas such as employment, essential private and public services, creditworthiness, and risk assessment and pricing in life and health insurance.

Whether your product is in scope is a legal question about your use case, not a technical one. Settle it before anything else on this page matters to you.