What questions does the AI Act assessment in Complipath ask?

The main-line questions

• Does this system make, support or inform a decision about a person?
• What is this system built and offered to do? The Annex III areas to tick, or none of these.
• Does the system do any of the following? The Article 5 practices that turn on a judgement, asked of every system.
• Is the output acted on automatically, or does a person decide? Asked only when an Annex III area was ticked.
• Can they realistically overrule it, and do they see the underlying case themselves? Asked when an area was ticked and the output is not acted on automatically.
• Does the system interact directly with people, or generate audio, image, video or text?
• Did you build it, fine-tune a third-party model, or buy it?

What it does

• Follow-ups open when an answer calls for one: for biometrics, law enforcement, employment, credit, critical infrastructure and generated content.
• Any question can be answered 'I don't know yet', or flagged for legal review. The assessment is saved with that question open, nothing is assumed and no classification is made until it is answered.
• The rule engine returns a risk level, the provision it rests on and the reasons. Each run is saved as a new assessment, with the engine version and the question wording; earlier ones are kept.
• A provision that has not started to apply is set aside as coming, with its date and sets no risk level.
• The result is a draft until a person confirms it. Confirming creates the requirements checklist.

What it does not do

• It does not let the questions on who acts on the output change the class. They are recorded as material for the provider's own Article 6(3) assessment.
• It never concludes that an Annex III system is not high-risk. Article 6(3) lets a provider reach that conclusion, and Article 6(4) requires the provider to document it.
• It works from your answers and does not inspect the system, so a wrong answer gives a wrong result.
• A language model does not set an answer. Suggestions are never pre-selected, and the database rejects anything but the engine's closed set of answer keys.

Which plan

Every plan, including the free first system.

Where it sits in the law

• Article 6(2) and Annex III: AI systems referred to in Annex III are considered high-risk, subject to the derogation in Article 6(3). The area question and its follow-ups map your answers to the Annex III points.
• Article 6(3): by derogation from Article 6(2), an Annex III system is not considered high-risk where it poses no significant risk of harm to the health, safety or fundamental rights of natural persons, including by not materially influencing the outcome of decision making.
• That applies where any of four listed conditions is fulfilled. An Annex III system that performs profiling of natural persons is always high-risk.
• Article 6(4): a provider who considers an Annex III system not high-risk must document the assessment before the system is placed on the market or put into service, and is subject to the registration obligation in Article 49(2). The assessment does not make that conclusion.
• Article 50: the answers on interaction, generated content and biometrics raise its transparency duties. Each of paragraphs 1 to 4 has its own exceptions; the engine states that a duty applies and leaves any exception to a person.