In this article7 sections
The Classification Question That Changes Everything
AI Act Article 6 high-risk classification is not about how sophisticated your AI tool is. It is about what the tool does and who it affects. Microsoft Copilot is a capable AI assistant, but whether it counts as a high-risk system under the EU AI Act depends entirely on how your organisation deploys it, not on the Microsoft brand name or the underlying model.
Get this classification wrong and you face two very different problems. Classify Copilot as high-risk when it is not, and you waste time and money on unnecessary controls. Miss a genuine high-risk deployment and, from 2 December 2027, breaching deployer duties can be fined up to €15 million or 3% of global annual turnover under Article 99, the lower of the two for SMEs. The test in Article 6 is how you find the right answer.
The Article 6 Test Explained
Article 6 of the EU AI Act has two routes to high-risk status. Route 1: safety components of products covered by Annex I that need a third-party conformity assessment (Art 6(1)). Those rules apply from 2 August 2028 and rarely concern Copilot. Route 2: uses listed in Annex III (Art 6(2)), which apply from 2 December 2027.
Route 2: Is the use listed in Annex III?
Annex III lists eight specific use-case areas:
-
Biometric identification and categorisation
-
Critical infrastructure management
-
Education and vocational training
-
Employment, workers management, and access to self-employment
-
Access to essential private and public services and benefits
-
Law enforcement
-
Migration, asylum, and border control management
-
Administration of justice and democratic processes
Category 4 is the one most SMEs need to watch. It explicitly covers AI used for recruitment, CV screening, promotion decisions, task allocation, and performance monitoring.
The narrow exception: Article 6(3).
An Annex III system escapes high-risk status only if it poses no significant risk of harm and fits one of four narrow conditions: it performs a narrow procedural task, improves the result of a human activity already completed, detects patterns without replacing human assessment, or only performs a preparatory task. The provider must document that assessment (Art 6(4)). A system that profiles people is always high-risk.
A Copilot deployment that drafts a generic project update email is not an Annex III use at all. A Copilot deployment configured to rank job applicants or flag employees for performance review is. If your use case does not appear in Annex III (and is not a product safety component), the system is not high-risk under Article 6, regardless of how impactful it seems internally.
A Practical Decision Tree for Copilot Deployments
Walk through these questions for each Copilot integration or plugin your organisation runs.
Step 1. What task is Copilot performing? Write down the concrete output. "Summarise emails" is different from "score candidate suitability."
Step 2. Does that task map to one of the eight Annex III categories? Be honest. "HR analytics" is not automatically Annex III Category 4. Only deployments that make or substantially influence decisions about employment qualify.
Step 3. Does one of the Article 6(3) exceptions clearly fit, and does the system avoid profiling people? Only then can an Annex III use fall outside high-risk. Ranking applicants or scoring employees is profiling, so it stays high-risk.
Step 4. Is there a meaningful human review before the decision is acted on? Human oversight does not change the classification, but for high-risk systems Article 26 requires it from 2 December 2027, and it must be real, not cosmetic. A tick-box approval of an AI-generated shortlist without any independent review does not count.
If the answers confirm a high-risk deployment, your Article 26 obligations apply from 2 December 2027.
Common Misconceptions That Get Companies Into Trouble
"Microsoft handles compliance, so we don't have to."
Microsoft is the provider. You are the deployer. Article 26 places a distinct set of obligations on deployers that no vendor can discharge on your behalf. For high-risk use, from 2 December 2027, you are responsible for using the system according to its instructions, assigning human oversight and keeping logs.
"Copilot is just a productivity tool, it can't be high-risk."
The AI Act does not classify tools. It classifies use cases. The same Copilot licence used to draft meeting summaries (not high-risk) could be configured via a plugin to analyse employee productivity patterns and flag underperformers (potentially high-risk under Annex III Category 4). The tool is identical. The deployment is not.
"We are too small to be regulated."
The AI Act applies to any organisation that deploys AI systems affecting people in the EU, regardless of headcount. The regulation does not have an SME exemption for deployers, though it does offer some procedural flexibilities for providers who are SMEs.
"We need a full AI impact assessment for everything."
A Fundamental Rights Impact Assessment (FRIA, Article 27) is only required for public bodies, private organisations providing public services, and deployers who use AI to assess the creditworthiness of individuals or to price life and health insurance. It applies from 2 December 2027. Most private SMEs never need one, but the Article 26 duties still apply to all high-risk deployers from that date.
Worked Examples by Use Case
Example 1: Marketing team using Copilot to draft campaign copy
Output: Text for email campaigns. No Annex III category applies. No rights or employment decisions involved. Result: Not high-risk. Standard AI literacy obligations under Article 4 apply, but nothing more.
Example 2: Finance team using Copilot to summarise contract clauses
Output: Summaries that a human lawyer reviews before any action. Annex III does not cover internal legal review tools used by private companies in a support capacity. Result: Not high-risk. Ensure staff understand the tool's limitations and that final decisions remain with qualified humans.
Example 3: HR team using a Copilot plugin that scores CVs and ranks applicants
Output: Ranked shortlists that recruiters use to decide who gets an interview. This squarely matches Annex III Category 4 (employment decisions) and affects fundamental rights (equal access to work). Result: High-risk. From 2 December 2027 you must implement human oversight and keep logs (Article 26). And if you configure a general-purpose assistant to score CVs yourself, you may become the provider of a high-risk system (Art 25(1)(c)), with provider obligations. Take advice before building this.
Example 4: Operations manager using Copilot to generate shift schedules
Output: Draft rosters that the manager adjusts before publishing. No automated disciplinary or dismissal decision involved. Human review is genuine. Result: Usually not high-risk. Annex III covers task allocation based on individual behaviour or personal traits. If the tool assigns shifts on that basis, revisit.
Example 5: Copilot integrated with a performance management platform, flagging employees below a KPI threshold for manager attention
Output: Automated alerts used to initiate performance improvement plans. This combination affects employment and is driven by automated analysis. Result: High-risk under Annex III Category 4. Article 26 obligations apply from 2 December 2027.
What Your IT Team Needs to Do Right Now
Deployer duties for high-risk systems (Article 26) apply from 2 December 2027 for Annex III uses such as recruitment. Use the time to classify now. Three actions matter most.
First, build an AI inventory. List every Copilot integration, plugin, and connected workflow. For each one, run the Article 6 test above. Document your reasoning. This record is what a regulator or a works council will ask for first.
Second, check your Microsoft agreement. Microsoft publishes documentation describing the intended purpose of each Copilot product. Article 13 requires providers of high-risk systems to give deployers instructions for use. If that documentation does not exist or is unclear, escalate to your Microsoft account manager in writing.
Third, brief your HR and department heads. The classification decision is not purely technical. It requires input from the people who know how outputs are actually used day-to-day. A monthly AI governance review meeting, even a short one, builds the habit your compliance team will thank you for later.
The EU AI Act does not punish organisations for using AI. It penalises organisations that use it without knowing what they are doing. Running the Article 6 test is how you demonstrate that your organisation does know what it is doing.
Not sure whether your Copilot deployment crosses the high-risk threshold? Run the free 2-minute compliance check at comply.khairos.ai and get a clear starting point for your AI Act obligations.
References
Not sure where your company stands?
Answer 15 questions about how your team uses AI. You get a compliance score and the gaps to close first, as a PDF.