← Back to FasterUp.AI
Governance of the AI agents
Autonomy levels
- T0 — observe and report: the agent only describes what it sees; no action.
- T1 — recommend: proposes the action, a human approves it (MLEO quarantine with supervisor approval).
- T2 — pre-approved playbook: actions defined and tested in advance, with a circuit-breaker: on an abnormal volume of decisions the agent stops itself.
- T3 — autonomous on an allowlist: only blocks of known indicators (IP, domain, hash), with automatic rollback on expiry.
The level is set per client and per action type; by default, actions with production impact stay at T1.
Safeguards
- Kill-switch: any client can stop automated response; the sensor stays in detection.
- Allowlist: agents act only on the target classes defined in the contract; the client's critical infrastructure is explicitly excluded.
- Immutable journal: every decision has a transaction ID, source, reason and time; the journal cannot be edited.
- Rollback by ID: any action is reversed by its transaction ID, in seconds.
- Never on the critical data path: AI components are not inline; if a model does not answer, traffic is unaffected.
Vendor independence
The platform uses several AI engines with automatic failover: no external vendor can stop the operations center. The CYBER3.AI security model runs in the European Union.
Compliance: aligned with the EU Artificial Intelligence Act requirements for systems under human oversight; details on request.
Write to us: [email protected]
Ecosystem
- CYBER3.AISecurity LLM and threat intelligence
- CYBER3Anti-scam apps, VPN included
- FasterUp SOCManaged SOC & VAS with sensors
- FasterUp OTIndustrial network and SCADA monitoring
All are brands of ROL PORTAL SERVICES SRL · ISO/IEC 27001:2022
© 2026 ROL PORTAL SERVICES SRL · CUI RO 36837313