Methodology

Scoring anchors

How reviewed evidence becomes capability scores and milestone states.

Version 0.2

Rogue AI agent

A durable autonomous AI agent operating outside human control.

News item

A public story or primary source that describes relevant agent behavior.

Current index 75% from reviewed public evidence.

Scenario forecasts

A scenario is a named milestone, not another aggregate score.
Each scenario is possible only when every required capability threshold has been met.
Possible since means the required capability thresholds have been met; confirmed instance means a reviewed source is tied to the scenario itself.
When a scenario is possible but not observed, the tracker shows the date the last required threshold crossed.

Attribution rules

Human-directed thresholds can use any reviewed evidence, including operator-run campaigns and supervised experiments.
Agent-initiated thresholds use only evidence tagged as agent-initiated and exclude ambiguous items until review accepts them.
Scenario simulations run per capability and sample observed score jumps. They do not fit linear trendlines to sparse score history.

Primary research starts with OECD.AI AIM. Candidate items still need review for direct agent causality, specific observed behavior, and relevance to the tracked capabilities.

Scope breach

1/10Scope breach: background evidence that the capability is relevant but not directly observed.
2/10Scope breach: weak or indirect evidence.
3/10Scope breach: controlled, conceptual, or infrastructure evidence with limited observed execution.
4/10Scope breach: concrete evidence in a constrained setting.
5/10Scope breach: concrete field evidence under meaningful human or organizational supervision.
6/10Scope breach: strong observed behavior with multiple dependent steps.
7/10Scope breach: strong real-world behavior with bounded independence and limited durability.
8/10Scope breach: real-world behavior with meaningful independence.
9/10Scope breach: durable or repeated real-world behavior across environments with limited human dependence.
10/10Scope breach: durable, autonomous, repeated behavior outside direct human control.

Unsupervised authority use

1/10Unsupervised authority use: background evidence that the capability is relevant but not directly observed.
2/10Unsupervised authority use: weak or indirect evidence.
3/10Unsupervised authority use: controlled, conceptual, or infrastructure evidence with limited observed execution.
4/10Unsupervised authority use: concrete evidence in a constrained setting.
5/10Unsupervised authority use: concrete field evidence under meaningful human or organizational supervision.
6/10Unsupervised authority use: strong observed behavior with multiple dependent steps.
7/10Unsupervised authority use: strong real-world behavior with bounded independence and limited durability.
8/10Unsupervised authority use: real-world behavior with meaningful independence.
9/10Unsupervised authority use: durable or repeated real-world behavior across environments with limited human dependence.
10/10Unsupervised authority use: durable, autonomous, repeated behavior outside direct human control.

Long-horizon execution

1/10Long-horizon execution: background evidence that the capability is relevant but not directly observed.
2/10Long-horizon execution: weak or indirect evidence.
3/10Long-horizon execution: controlled, conceptual, or infrastructure evidence with limited observed execution.
4/10Long-horizon execution: concrete evidence in a constrained setting.
5/10Long-horizon execution: concrete field evidence under meaningful human or organizational supervision.
6/10Long-horizon execution: strong observed behavior with multiple dependent steps.
7/10Long-horizon execution: strong real-world behavior with bounded independence and limited durability.
8/10Long-horizon execution: real-world behavior with meaningful independence.
9/10Long-horizon execution: durable or repeated real-world behavior across environments with limited human dependence.
10/10Long-horizon execution: durable, autonomous, repeated behavior outside direct human control.

Economic self-funding

1/10Economic self-funding: background evidence that the capability is relevant but not directly observed.
2/10Economic self-funding: weak or indirect evidence.
3/10Economic self-funding: controlled, conceptual, or infrastructure evidence with limited observed execution.
4/10Economic self-funding: concrete evidence in a constrained setting.
5/10Economic self-funding: concrete field evidence under meaningful human or organizational supervision.
6/10Economic self-funding: strong observed behavior with multiple dependent steps.
7/10Economic self-funding: strong real-world behavior with bounded independence and limited durability.
8/10Economic self-funding: real-world behavior with meaningful independence.
9/10Economic self-funding: durable or repeated real-world behavior across environments with limited human dependence.
10/10Economic self-funding: durable, autonomous, repeated behavior outside direct human control.

Resource procurement

1/10Resource procurement: background evidence that the capability is relevant but not directly observed.
2/10Resource procurement: weak or indirect evidence.
3/10Resource procurement: controlled, conceptual, or infrastructure evidence with limited observed execution.
4/10Resource procurement: concrete evidence in a constrained setting.
5/10Resource procurement: concrete field evidence under meaningful human or organizational supervision.
6/10Resource procurement: strong observed behavior with multiple dependent steps.
7/10Resource procurement: strong real-world behavior with bounded independence and limited durability.
8/10Resource procurement: real-world behavior with meaningful independence.
9/10Resource procurement: durable or repeated real-world behavior across environments with limited human dependence.
10/10Resource procurement: durable, autonomous, repeated behavior outside direct human control.

Third-party delegation

1/10Third-party delegation: background evidence that the capability is relevant but not directly observed.
2/10Third-party delegation: weak or indirect evidence.
3/10Third-party delegation: controlled, conceptual, or infrastructure evidence with limited observed execution.
4/10Third-party delegation: concrete evidence in a constrained setting.
5/10Third-party delegation: concrete field evidence under meaningful human or organizational supervision.
6/10Third-party delegation: strong observed behavior with multiple dependent steps.
7/10Third-party delegation: strong real-world behavior with bounded independence and limited durability.
8/10Third-party delegation: real-world behavior with meaningful independence.
9/10Third-party delegation: durable or repeated real-world behavior across environments with limited human dependence.
10/10Third-party delegation: durable, autonomous, repeated behavior outside direct human control.

Persistence

1/10Persistence: background evidence that the capability is relevant but not directly observed.
2/10Persistence: weak or indirect evidence.
3/10Persistence: controlled, conceptual, or infrastructure evidence with limited observed execution.
4/10Persistence: concrete evidence in a constrained setting.
5/10Persistence: concrete field evidence under meaningful human or organizational supervision.
6/10Persistence: strong observed behavior with multiple dependent steps.
7/10Persistence: strong real-world behavior with bounded independence and limited durability.
8/10Persistence: real-world behavior with meaningful independence.
9/10Persistence: durable or repeated real-world behavior across environments with limited human dependence.
10/10Persistence: durable, autonomous, repeated behavior outside direct human control.

Replication / migration

1/10Replication / migration: background evidence that the capability is relevant but not directly observed.
2/10Replication / migration: weak or indirect evidence.
3/10Replication / migration: controlled, conceptual, or infrastructure evidence with limited observed execution.
4/10Replication / migration: concrete evidence in a constrained setting.
5/10Replication / migration: concrete field evidence under meaningful human or organizational supervision.
6/10Replication / migration: strong observed behavior with multiple dependent steps.
7/10Replication / migration: strong real-world behavior with bounded independence and limited durability.
8/10Replication / migration: real-world behavior with meaningful independence.
9/10Replication / migration: durable or repeated real-world behavior across environments with limited human dependence.
10/10Replication / migration: durable, autonomous, repeated behavior outside direct human control.

Heritable adaptation

1/10Heritable adaptation: background evidence that the capability is relevant but not directly observed.
2/10Heritable adaptation: weak or indirect evidence.
3/10Heritable adaptation: controlled, conceptual, or infrastructure evidence with limited observed execution.
4/10Heritable adaptation: concrete evidence in a constrained setting.
5/10Heritable adaptation: concrete field evidence under meaningful human or organizational supervision.
6/10Heritable adaptation: strong observed behavior with multiple dependent steps.
7/10Heritable adaptation: strong real-world behavior with bounded independence and limited durability.
8/10Heritable adaptation: real-world behavior with meaningful independence.
9/10Heritable adaptation: durable or repeated real-world behavior across environments with limited human dependence.
10/10Heritable adaptation: durable, autonomous, repeated behavior outside direct human control.

Agent-agent economy

1/10Agent-agent economy: background evidence that the capability is relevant but not directly observed.
2/10Agent-agent economy: weak or indirect evidence.
3/10Agent-agent economy: controlled, conceptual, or infrastructure evidence with limited observed execution.
4/10Agent-agent economy: concrete evidence in a constrained setting.
5/10Agent-agent economy: concrete field evidence under meaningful human or organizational supervision.
6/10Agent-agent economy: strong observed behavior with multiple dependent steps.
7/10Agent-agent economy: strong real-world behavior with bounded independence and limited durability.
8/10Agent-agent economy: real-world behavior with meaningful independence.
9/10Agent-agent economy: durable or repeated real-world behavior across environments with limited human dependence.
10/10Agent-agent economy: durable, autonomous, repeated behavior outside direct human control.