A green–amber–red status can show whether an AI system is operating normally, needs attention or has stopped. It cannot describe what an agent may do, how serious a failure would be, who checks the action or who remains accountable.
Methodfield has published a new bilingual analysis: Multi-stage AI-agent control: authority, accountability and independent oversight.
Six coordinates replace one trust colour
The guide introduces an A · C · S · I · R · M system card:
- authority;
- consequence criticality;
- supervision depth;
- independence of control paths;
- accountability allocation;
- current operating mode.
Traffic-light colours remain useful, but only for the current mode. They no longer stand in for a permanent level of trust assigned to an agent.
Control is both multi-level and multi-stage
The central working diagram is an L0–L5 × T0–T6 matrix.
Its rows distinguish agent self-check, deterministic maths and rules, independent AI verification, AI risk coordination, accountable human authority, and system assurance. Its columns cover change admission, goal, plan, pre-execution review, execution, observation and recovery.
This makes gaps visible. A final approval cannot compensate for missing identity checks, an uncontrolled executor or the absence of a recovery path.
Use the right mechanism at each point
The guide also separates four different technical roles:
- exact calculations and invariants belong in ordinary code and mathematics;
- stable probabilistic patterns may belong in classical ML;
- ambiguous language, evidence and trajectories may justify an AI controller;
- consequential exceptions remain with an accountable person.
Control agents receive asymmetric power: veto, request evidence, narrow scope or escalate. They do not receive permission to expand authority or execute a disputed action themselves.
Connected to the Knowledge Map
The new guide now appears as an AI Systems node in the Methodfield Knowledge Map. Editorial relations connect it to the existing guides on agent authority, automation types, staged autonomy and post-launch ownership.
The release includes a dedicated visual system: a new hero, localised six-coordinate panels, a bilingual control matrix and an illustration of independent redundancy.