The AI Act sets duties for providers and deployers of AI systems by risk level. Buyers of AI tools are usually deployers: they need AI literacy, must use high-risk systems as instructed, and depend on vendor documentation to meet their own duties.
Who it applies to
Providers that place AI systems or general-purpose AI models on the EU market, and deployers that use AI systems professionally in the EU, wherever they are based. Most organisations that buy AI tools are deployers.
Key dates
- 1 Aug 2024Entered into force
- 2 Feb 2025Prohibited practices (Art. 5) and AI literacy (Art. 4) apply
- 2 Aug 2025Obligations for general-purpose AI models apply
- 2 Aug 2026Transparency obligations (Art. 50) apply
- 2 Dec 2027High-risk systems in Annex III, after the Digital Omnibus amendment
- 2 Aug 2028High-risk systems that are safety components of Annex I products
What EU AI Act asks of third-party risk management
Providers and deployers must ensure a sufficient level of AI literacy among staff who use AI systems on their behalf.
Certain uses are banned outright, such as social scoring and some forms of emotion recognition at work.
High-risk systems come with instructions for use that explain capabilities, limits and human oversight.
A deployer that puts its name on a high-risk system or substantially modifies it can become its provider.
Use high-risk systems as instructed, assign human oversight, monitor operation, keep logs and inform affected workers.
People must know when they interact with AI, and synthetic content must be marked.
Model providers keep technical documentation, give downstream providers information and publish a training data summary.
Documents to request from vendors
- Description of the AI functionality and the models used
- List of model providers, also on the sub-processor list
- Data use terms: whether your data trains models
- Instructions for use and human oversight description, for high-risk systems
- Risk classification of the system under the AI Act
- Evaluation or testing results for accuracy and bias
- Transparency and content marking approach
In ProcureCortex
ProcureCortex reads AI vendors' documentation, DPAs and sub-processor lists against AI Act and AI policy rules you select, such as model providers missing from the sub-processor list or a default opt-in to training on your data.
How an assessment worksSources
- Regulation (EU) 2024/1689 (AI Act)eur-lex.europa.eu/eli/reg/2024/1689/oj
Summarised for practitioners, not legal advice. Check the source text for your own obligations.