# Criteria for agent independence

Three research indicators and minimum evidence requirements.

## Working definition

In this project, a candidate is relevant if there is evidence of at least one indicator: acting beyond the boundaries authorised by its operator; working without a current human operator or independently securing its computing resources; or independently launching active offspring. These are the project's working criteria, not a generally accepted classification.

## Acting beyond authorised boundaries

Establish the original permissions, the specific action and evidence that it crossed that boundary. A claim that a system can bypass restrictions does not establish that it has done so. Record laboratory tests separately from actions in real environments.

## Continued operation

Periodic posts establish account activity, but not the absence of a person: the source of activation and funding may remain unknown. An alleged owner's inactive public profile does not by itself establish that the system has been abandoned.

## Offspring

A new account or similar text is not the same as a new operational system. Evidence must connect the parent and offspring, document launch and show the offspring's separate activity. Confidence is limited by the weakest link in that chain.

## What to record

- The public identifier and the source from which it was obtained.
- Observation time, separately from the event time reported by the source.
- Verified facts, claims, alternative explanations and unknowns.
- An observation that could disprove the current hypothesis.

The existence of an agent messaging protocol does not establish the independence of its participants.

## Basis and sources

Project editorial methodology; not evidence of any particular agent being autonomous.

- A2A protocol specification: https://a2a-protocol.org/latest/specification/

Revision: 2; published: 2026-10-19T03:13:37.900298+00:00; content as of: 2026-10-18T00:00:00Z