ISO 42001
Put AI governance, impact assessment and lifecycle controls into a certifiable management system.
Artificial Intelligence Act · Regulation (EU) 2024/1689
Classify your AI systems and map obligations to your role and the current phase-in dates.
Providers, deployers and other actors within the Act’s territorial and material scope, including certain actors outside the EU.
Role- and risk-specific governance, transparency, documentation and conformity evidence as applicable.
Separate duties for general-purpose AI models, specific transparency cases and high-risk systems. Check the current amended timetable and sector-specific routes.
Track unclassified AI uses, applicable deadlines, failed oversight tests, missing provider documentation and systems approved for one purpose but used for another.
Fill this in with your sponsor and send it to the people who own the work.
EU AI Act kickoff brief Business reason and required outcome [Contract, obligation or approval] Scope and exclusions [Entities, systems or activities, and reasons] Fixed deadlines and internal milestones [Date, source and owner] Executive sponsor and authorized signer [Names] Project lead and backup [Names] First workstream Identify AI systems, models and actor roles · AI governance lead + legal [Name and due date] People, effort and external costs [Estimate, owner and contingency] Evidence folder and reviewer access [Location, access owner and review date] Decisions and blockers [Item, options, owner and decision date] First week 1. Have Procurement, HR, Engineering and products list active and planned AI uses, including vendor tools and bundled features. 2. Pick a consequential use. Trace the user, model and version, decision influenced, intervention available and role of each organization. 3. Give Legal and the operator a clear classification question, preferred owner and date. Put unresolved uses on the next release review agenda.
The phased rollout began with prohibited practices and the original AI literacy provisions. The AI Omnibus subsequently revised the literacy provisions.
Official reference (opens in a new tab)Check rules for models already on the market and current transition dates.
Official reference (opens in a new tab)The AI Omnibus extended the schedule for Annex III high-risk systems.
Official reference (opens in a new tab)Check how the amended schedule and sector-specific conformity route apply to your product.
Official reference (opens in a new tab)Checklist, instructions and 12 templates.
| Step | Owner | When to act | Estimated target | Deliverables | Status |
|---|---|---|---|---|---|
| Identify AI systems, models and actor roles | AI governance lead + legal | At initiation | Set a start date | Inventory; Actor-role analysis; Scope decision | Open |
| Classify and assign obligations | Product counsel + risk lead | Before development and market placement | Set a start date | Classification memo; Obligation register; Phase-in schedule | Open |
| Implement controls and evidence | Model owners + data governance | Before applicable obligations take effect | Set a start date | Training evidence; Technical documentation; Evaluation and oversight records | Open |
| Complete required assessments and monitor | Conformity lead + operations | Before market placement and ongoing | Set a start date | Assessment file; Monitoring plan; Incident workflow | Open |
Targets use illustrative calendar-day estimates and assume the steps run in sequence. Check official dates and agree real durations with your team.
Inventory uses and model dependencies, intended purposes, market jurisdictions and provider/deployer responsibilities.
Product specifications, vendors, deployment geography, intended purposes, version history and contracts.
An applicant-screening entry identifies the end use, decision influenced, version, vendor, inputs, people affected and who may override the output.
Escalate shared or rebranded systems, unclear supplier responsibilities or a product change with no role review.
Fictional details. Use the same fields with evidence from your work.
Each item has one accountable owner. The source and as-of date let a reviewer trace it back.
Share the reviewed inventory with the people building the control, privacy or testing scope.
Fictional details. Use the same fields with evidence from your work.
A reviewer can reach the same decision using the source and facts, or see exactly what remains unresolved.
Send confirmed obligations to the delivery team and record any unresolved decision with the appropriate authority or specialist.
Fictional details. Use the same fields with evidence from your work.
A reviewer can use the memo to classify a named system and see how shared dependencies are handled.
Give the approved scope and any open questions to whoever will map requirements, select samples or review the engagement.
Screen prohibited practices, high-risk criteria, transparency triggers and any general-purpose model obligations. Map applicability and dates by system.
Approved inventory, factual use, affected groups, current rules and applicable sector legislation.
The classification memo shows what the system actually does, which provisions were checked, why they apply and the next review trigger.
Bring unresolved prohibited-use or high-risk questions to Legal before launch and escalate missing documentation for a role with obligations.
Fictional details. Use the same fields with evidence from your work.
A reviewer can reach the same decision using the source and facts, or see exactly what remains unresolved.
Send confirmed obligations to the delivery team and record any unresolved decision with the appropriate authority or specialist.
Fictional details. Use the same fields with evidence from your work.
A reviewer can reach the same decision using the source and facts, or see exactly what remains unresolved.
Send confirmed obligations to the delivery team and record any unresolved decision with the appropriate authority or specialist.
Fictional details. Use the same fields with evidence from your work.
Every external deadline links to a real source. Planning targets are visibly marked.
Add approved dates to the team calendar and share them with the people scheduled to deliver or review.
Build required risk and data governance, oversight, evaluations, logging and documentation. Implement literacy and user transparency where applicable.
Applicable obligations, technical documentation, evaluation data, oversight design, instructions and product release plan.
An oversight test records a harmful or incorrect output, the reviewer’s intervention, what the downstream user saw and whether action completed before the decision.
Raise an untested override, a material performance gap or a system operating beyond its approved purpose.
Fictional details. Use the same fields with evidence from your work.
A reviewer can open the right file, reproduce the population filter and see who performed or approved the real task.
Send the index and access instructions to the assessor. Keep sensitive files in the approved evidence store.
Fictional details. Use the same fields with evidence from your work.
The submitted package can be recreated from the index and the receipt is linked to the exact files.
Send the approved package through the official channel. Treat the download as an index, and use issuer forms where required.
Fictional details. Use the same fields with evidence from your work.
An absent participant can see what was tested, what happened and why an action was assigned.
Send approved findings to action owners and file the record for governance or assessment.
Select the relevant conformity route where required. Track declarations, registrations, incidents and substantial modifications.
Classification, evaluation results, provider materials, relevant standards and applicable release conditions.
A dated release record links the completed applicable assessment, version, evidence, monitoring thresholds and named incident contact.
Escalate failed release conditions, suspected serious incidents or a supplier change that breaks the evidence used for approval.
Fictional details. Use the same fields with evidence from your work.
The submitted package can be recreated from the index and the receipt is linked to the exact files.
Send the approved package through the official channel. Treat the download as an index, and use issuer forms where required.
Fictional details. Use the same fields with evidence from your work.
A person who receives the plan can tell when to act, what to check and where to save the result.
Give the approved plan to the people responsible for operation. Add dated checkpoints to the working calendar.
Fictional details. Use the same fields with evidence from your work.
Every decision has one accountable person, a backup and a practical route to reach them.
Share the approved instructions with the people expected to follow them and test the handoffs at the next exercise.