C7: Safety & Security Model Reports

Document model capabilities, risks, and mitigations for the AI Office.

Step 7 of 10

Commitment Progress0% (0/9)

REQ-7.1-01Measure 7.1 (Page 19)

Does the Model Report include a description of architecture, training data, and capabilities?

"Signatories will provide in the Model Report: a high-level description of the model's architecture, capabilities, propensities... training method and data."

REQ-7.1-02Measure 7.1 (Page 19)

Are intended use, restrictions, and key benchmark performance documented?

"description of the intended use and known use restrictions... and the model's performance on key benchmarks."

REQ-7.2-01Measure 7.2 (Page 19)

Are adversarial capabilities and propensities reported?

"Signatories will provide... the model's capabilities... relating to systemic risks... and the model's propensity to facilitate threat actors' activities."

REQ-7.3-01Measure 7.3 (Page 20)

Does the Model Report document the results of systemic risk identification and analysis?

"Signatories will provide in the Model Report: (1) a description of the results of their systemic risk identification and analysis... including... (d) a description of the systemic risks stemming from the model and a justification therefor."

REQ-7.3-02Measure 7.3 (Page 20)

Are model evaluation results and methods fully described?

"Signatories will provide... (e) all results of model evaluations relevant to understanding the systemic risks... and descriptions of: (i) how the evaluations were conducted..."

REQ-7.4-01Measure 7.4 (Page 21)

Are reports from independent external evaluators included?

"Signatories will provide in the Model Report: (1) any available reports (e.g. via valid hyperlinks) from: (a) independent external evaluators involved in model evaluations..."

REQ-7.5-01Measure 7.5 (Page 21)

Does the report address material changes to the systemic risk landscape?

"Signatories will ensure that the Model Report contains information relevant for the AI Office to understand whether and how the development... result in material changes in the systemic risk landscape..."

REQ-7.6-01Measure 7.6 (Page 22)

Is the Model Report updated when risk justifications are undermined?

"Signatories will update their Model Report if they have reasonable grounds to believe that the justification for why the systemic risks... are acceptable... has been materially undermined."

REQ-7.7-01Measure 7.7 (Page 22)

Is the Model Report provided to the AI Office before market placement?

"Signatories will provide the AI Office with access to the Model Report... by the time they place a model on the market."