Skip to content

Autonomous Agent (Generic)

Why this matters now

Agentic systems are being deployed across low‑ and high‑risk domains. They can act without immediate human approval, which makes delegation, accountability, and intervention the core safety problems.

GTAF is designed to make those problems explicit and verifiable without prescribing tools.

What GTAF does for agentic systems

  • Defines the boundary where the agent is allowed to act (SB).
  • Defines the decision space and its limits (DR).
  • Binds outcome ownership to a human/organizational role (RB).
  • Gates activation with a binary readiness check (DRC).
  • Scales strictness by risk class (A/B/C).

Minimal adoption (low‑risk agent)

If you deploy an agent in a bounded, reversible domain (Class A):

  1. Create SB (what is in/out).
  2. Define DR (what the agent may decide).
  3. Bind RB (who owns outcomes).
  4. Issue DRC (PERMITTED or NOT_PERMITTED).

This is enough to make delegation explicit and auditable, without heavy governance.

When risk increases (Class B/C)

As impact, irreversibility, or autonomy increases:

  • DRB becomes mandatory (drift review).
  • EIS becomes mandatory (intervention/kill‑switch).
  • DVM becomes mandatory (early warning signals).

Agentic Delegation Flow (non-binding)

flowchart TB
  A[Agent introduced] --> B[Define SB]
  B --> C[Define DR]
  C --> D[Bind RB]
  D --> E[Evaluate DRC]
  E -->|PERMITTED| F[Delegation allowed]
  E -->|NOT_PERMITTED| G[Fix gaps]

Notes - The same structure applies across low‑ and high‑risk agents; only strictness changes. - If any required artifact is invalid, delegation is NOT_PERMITTED.

Start here