AI Feature Toggles

How AI Inference, AI Agents, AI Autonomy, and AI Training relate, who can change them, and what each flag actually gates.

The top of Settings → AI shows four switches: AI Inference, AI Agents, AI Autonomy, and AI Training. They apply to the whole account, not to individual users. Praetorian manages them: you can see each switch's state, but the switch is disabled with the tooltip Managed by Praetorian. Contact your Praetorian team to change one.

These switches do not choose models, set spend, or configure hunt profiles. They turn whole classes of AI behavior on or off for the account.

The four switches

AI Inference: "Human in the loop. Answers questions using your data; you act on the results." When on, the AI Assistant can use your account data to answer questions.

AI Agents: "Human on the loop. Acts on your behalf while you supervise and can intervene." When on, the AI Assistant can take actions such as triaging vulnerabilities and validating seeds. It is off until enabled, and it requires AI Inference. While Inference is off, the switch shows Enable AI Inference first.

AI Autonomy: "Human off the loop. Fully autonomous penetration testing via Hannibal." When on, admin and analyst users on the account can see hunt sessions. Users with the read-only role do not see hunts even while it is on. It is off until enabled, and it does not depend on AI Inference or AI Agents.

AI Training: when on, account data may be used to improve AI model performance under Praetorian's data-handling policies. It requires AI Agents. While Agents is off, the switch shows Enable AI Agents first.

Dependencies

AI Inference → AI Agents → AI Training form a chain. AI Autonomy stands on its own.

Turning off a switch also turns off everything that depends on it, so dependents do not come back on by themselves later. For example, turning off AI Inference also turns off AI Agents and AI Training.

Turning a switch on asks for confirmation first. Turning one off happens immediately.

AI Model Providers, Monthly AI Spend, and Automatic Penetration Tests are separate cards on the same tab and stay visible when a switch is off. The Automatic Penetration Tests card is locked with Enable AI Autonomy first while AI Autonomy is off.