EU AI Act High-Risk Systems for German Companies (2026)
Short answer
For German companies, EU AI Act high-risk systems are not one single bucket. As of July 9, 2026, the main 2 December 2027 date points to stand-alone Annex III systems such as recruitment, worker management, biometric, education, and essential-services scoring tools, while AI embedded in regulated products under Annex I follows the later 2 August 2028 branch in the current official implementation path.
- Annex III covers stand-alone high-risk use cases such as employment, biometrics, education, and scoring.
- Annex I covers AI embedded in regulated products such as certain medical devices, machinery, lifts, or toys.
- 2 December 2027 is the current official date to track for stand-alone Annex III systems.
- German buyers should classify the system and secure provider evidence before rollout, not in 2027.
EU AI Act high-risk systems for German companies fall into two different legal branches, and that is what changes the date answer. As of July 9, 2026, the main date to track for stand-alone Annex III high-risk AI systems is 2 December 2027, while AI embedded in regulated products under Annex I follows the later 2 August 2028 branch in the current official implementation path. For German companies, the practical question is therefore not only “Is this high-risk AI?” but also “Is this an Annex III software use case or an Annex I product case?”
That distinction matters immediately for procurement. A recruitment-screening platform, worker-scoring tool, biometric verification workflow, or education-assessment system is usually a different compliance category from AI embedded in a medical device, machinery, lift, or toy. If a German company mixes those categories, it can build the wrong contract package, request the wrong evidence from the vendor, and prepare for the wrong deadline.
For the wider date map, start with our EU AI Act timeline for Germany. For the contract layer underneath it, read EU AI Act procurement requirements for German enterprises. For the privacy workstream that often runs in parallel, see GDPR AI procurement.
What counts as high-risk AI under the EU AI Act?
Under Article 6 of the AI Act, high-risk AI systems split into two main categories.
Annex I high-risk AI: product-embedded systems
The first category covers AI that is used as a safety component of a product, or that is itself a product, under certain EU product-law frameworks. This is the Annex I branch. In practice, that means the AI sits inside a regulated product environment and is tied to the product’s conformity logic.
For German businesses, common examples can include AI integrated into:
- certain medical devices
- machinery or industrial equipment
- lifts
- toys
- other products that sit inside EU harmonisation legislation and conformity procedures
This is not the usual SaaS procurement case. The legal file tends to involve product documentation, manufacturer cooperation, conformity assessment pathways, and supplier change-control issues as well as AI Act obligations.
Annex III high-risk AI: stand-alone sensitive use cases
The second category covers AI whose high-risk status comes from the use case itself. This is the Annex III branch. These systems are usually stand-alone software, APIs, or vendor-managed workflows rather than physical products.
For German companies, the most commercially relevant Annex III areas are:
- employment and worker management, such as applicant filtering, ranking, promotion support, or termination support
- biometrics, including identification or verification in sensitive contexts
- education and vocational training, where AI influences admissions, evaluation, or progression
- access to essential private or public services, including certain scoring or eligibility decisions
- critical infrastructure
- migration, asylum, and border control
- administration of justice and democratic processes
This is why the keyword eu ai act high risk systems usually maps more closely to Annex III buyer-side questions than to general AI policy commentary. The searcher is often trying to work out whether a specific software deployment lands in the 2027 branch.
What changes on 2 December 2027?
For German companies, 2 December 2027 is now the date to track for stand-alone Annex III high-risk AI systems in the current official implementation framing. The European Commission’s July 2026 materials now separate that category from product-embedded systems, which are placed on the later 2 August 2028 line.
That does not mean every AI compliance obligation sleeps until the end of 2027. It means the stand-alone Annex III high-risk rule set should be planned around that date. Procurement, governance, classification, and evidence gathering still start much earlier.
The short operational answer for German companies
If your business is buying or deploying AI software for recruitment, worker management, biometric verification, educational assessment, or essential-services scoring, 2 December 2027 is now the most important high-risk date to track. If the AI is instead part of a regulated product, the relevant high-risk branch is the later 2 August 2028 line.
That is the snippet-ready answer. The longer legal answer is that the enacted AI Act timetable and the newer official implementation path still need to be read together. For practical business use, however, German procurement and legal teams now need the Annex III versus Annex I split first.
Annex III versus Annex I: the practical classification test
German companies do not need a perfect legal memo before asking the right questions. A useful first-pass classification test is:
- Is the AI being bought as software, API, or workflow tooling?
- Is the intended purpose one of the listed Annex III use cases?
- Or is the AI integrated into a regulated product environment that follows product-law conformity rules?
If the answer is “software plus an Annex III use case”, you are usually looking at the 2 December 2027 branch. If the answer is “embedded in a regulated product”, you are usually looking at the 2 August 2028 branch.
Typical Annex III procurement examples in Germany
The following cases often belong in the stand-alone Annex III discussion:
- AI for CV screening or candidate ranking
- AI for employee scoring or performance-related evaluation support
- biometric verification in access or identity workflows
- AI that influences student assessment or admissions
- AI used to score natural persons for access to credit, insurance, housing, or other essential services
These are exactly the kinds of tools where the deployer is a company in Germany and the provider is an external vendor. That makes the procurement file central.
Typical Annex I product-side examples
The following cases point toward the Annex I side:
- AI embedded in a medical product
- AI built into industrial machinery
- AI integrated into other equipment subject to product-law conformity assessment
These cases are less about ordinary software procurement and more about supplier documentation, manufacturer obligations, and product-law alignment.
Do all high-risk AI rules start in 2027?
No. For German companies, that statement is too broad and will usually mislead the business.
Why the “everything starts in 2027” shortcut is wrong
Three different timing layers matter:
- 2 August 2026 still matters for the broader AI Act framework and transparency-related rollout work.
- 2 December 2027 is the main date now associated with stand-alone Annex III high-risk AI systems in the current official implementation path.
- 2 August 2028 is the later date associated with high-risk AI embedded in regulated products.
That means the correct internal answer is not “the AI Act high-risk deadline is 2027”. The correct answer is “the date depends on whether the system is an Annex III stand-alone use case or an Annex I product case, and procurement work begins before either.”
For the broader date split, our timeline page gives the full 2026, 2027, and 2028 map. For the procurement-before-deadline question, see Does EU AI Act procurement matter before 2027?.
Provider and deployer duties: why German buyers should care now
Most German businesses buying enterprise AI are deployers. The vendor is usually the provider. That role split is one of the most important practical points in EU AI Act procurement.
What the provider usually controls
The provider typically controls:
- the system’s intended-purpose framing
- technical and compliance documentation
- instructions for use
- product or system change information
- incident reporting and remediation information
If the provider does not contractually commit to support the deployer, the buyer may have no clean path to operationalise its obligations later.
What the deployer needs to organise internally
A German company acting as deployer usually needs to be able to show:
- that the system is used consistently with the vendor’s instructions for use
- that human oversight is assigned internally
- that the business can monitor the tool and escalate issues
- that relevant input data and operational use are aligned with the intended purpose
- that privacy, labor-law, and sector-specific controls have been mapped where relevant
This is where the EU AI Act and GDPR often overlap. An HR tool may sit in an Annex III employment category while also raising Article 35 GDPR, Section 26 BDSG, or BetrVG issues. For that overlap, our GDPR AI procurement guide is the right companion page.
Procurement diligence questions for German companies
The most useful procurement question is not “Is the vendor AI Act compliant?” That question is too vague. The useful questions are narrower and evidence-based.
Questions to ask before contract signature
German legal, procurement, and compliance teams should ask:
- Does the vendor classify the system as Annex III, Annex I, or neither?
- What intended purpose does the vendor define in writing?
- What instructions for use will the deployer receive?
- What human-oversight assumptions does the system rely on?
- What logging or traceability support is available?
- How will the vendor notify the customer about serious incidents, updates, or scope changes?
- What documentation can the customer request or inspect?
- How do GDPR, data transfers, and employee-data issues interact with this deployment?
Those questions create a stronger procurement record than generic claims such as “AI Act ready” or “EU hosted”.
Contract points that should be settled early
Before rollout, German companies should usually secure:
- documentation access rights
- clear use instructions
- oversight and escalation support
- incident cooperation clauses
- change-management notices
- privacy and transfer terms where personal data is involved
For the fuller contract drafting layer, go to EU AI Act procurement requirements for German enterprises and GDPR AI procurement.
Need help classifying an EU AI Act high-risk system?
Compound Law advises German companies on AI system classification, procurement contracts, deployer obligations, GDPR overlap, and the distinction between stand-alone Annex III software and product-embedded AI. This page provides general information only and does not replace legal advice for a specific tool, product, or deployment.