The Commission's GPAI guidelines clarify which models are in scope and outline thresholds, documentation, and safety expectations, including criteria for systemic-risk models. This helps buyers and providers decide when EU AI Act General-Purpose AI duties apply.
Why this matters
Clearer scope and thresholds reduce classification disputes and accelerate procurement checks. The guidance supports uniform interpretation across Member States.
Impact on obligations
Provider
Assess whether the model meets the defined resource thresholds and prepare documentation accordingly.
Deployer
Update supplier questionnaires to reflect the guidelines.
Importer
Use the thresholds to screen incoming models before distribution in the EU.
What to evidence
- Scope assessment memo
- Threshold calculations
- Model factsheet
- Transparency-log inclusion proofs
Key artefacts explained:
- DSSE: Dead Simple Signing Envelope — portable signature format
- STH: Signed Tree Head — tamper-evident checkpoint in transparency log
- TSA: Time-Stamp Authority — independent timestamp receipt
- WORM: Write Once Read Many — immutable storage for audit trails