Self-assessment has two roles in AI regulation: a provider may argue that an Annex III system does not pose significant risk and should not be classified as high-risk; many high-risk providers also assess compliance internally, without a pre-launch review by an external body. section 7.2.3 section 7.3.3
Evidence for an Article 6(3) exception may include a system that:
- Performs narrow procedural tasks. section 7.2.3
- Improves an activity already completed by a human. section 7.2.3
- Detects decision patterns without influencing outcomes. section 7.2.3
- Prepares a summary that a human reviews independently before acting. section 7.2.3
These indicators are evidence, not automatic exemptions. For ClaimAssist, an adjuster checks the generated summary against source documents and decides independently. If the adjuster instead trusts the summary by default, automation bias weakens the case; profiling also rules out this self-assessment option. section 7.2.3 section 7.2.3
For internal compliance, the assessed system covers the components working together for its intended purpose, not just the predictive model. Providers must prepare technical documentation before launch and keep it current, ready to demonstrate compliance to market surveillance authorities. section 7.3.3 section 7.3.3


