Schwarz Backs Cohere: What Aleph Alpha Customers Do
Benedikt Langer
The Schwarz Group leads Cohere's funding round. At the same time, Cohere describes its North platform ...
The EU AI Act is the world’s first comprehensive AI law – and after transitional periods, 2026 becomes the year of enforcement. For companies, this means: the era of theoretical preparation is over. Implementation is now mandatory.
The good news: Not every company needs a compliance programme on the scale of a pharmaceutical giant. The AI Act is risk-based – the effort required scales with the risk category of each AI system deployed. The bad news: Most companies don’t even know which AI systems they’re using.
The AI Act classifies AI systems into four risk categories:
Unacceptable Risk: Social scoring, real-time remote biometric identification in publicly accessible spaces, and AI systems exploiting human vulnerabilities. These bans entered into force in February 2025.
High-Risk: AI used in areas such as recruitment, credit scoring, education, justice, critical infrastructure, and safety components. Here, the most stringent requirements apply: conformity assessment, risk management systems, data governance, transparency, human oversight, and technical documentation.
Limited Risk: Chatbots, deepfakes, emotion recognition systems. Primary obligations here are transparency-related: users must be informed that they are interacting with AI.
Minimal Risk: Spam filters, AI-assisted text processing, recommendation algorithms. No specific legal requirements apply – but voluntary codes of conduct are recommended.
The AI Act applies not only to AI developers but to the entire value chain:
Providers: Companies developing and placing AI systems on the market. They bear the primary compliance burden – including conformity assessments and CE marking for high-risk systems.
Deployers: Companies using AI systems. They must ensure high-risk systems are used strictly according to instructions, guarantee human oversight, and carry out fundamental rights impact assessments.
Importers and Distributors: Entities bringing AI systems from third countries into the EU must verify that those systems comply with the AI Act.
Most companies fall under the “deployer” category – and significantly underestimate their compliance obligations. Any company using AI-powered recruitment tools, automated credit scoring, or AI-based quality control is subject to deployer duties for high-risk systems.
Step 1: Build an AI Inventory. Which AI systems are currently in use? The answer almost always surprises. From automated invoice processing and AI-assisted hiring to customer service chatbots – many systems incorporate AI without it being obvious.
Step 2: Risk Categorisation. Each identified system is classified according to the AI Act’s risk-tier model. For high-risk systems, the rigorous compliance process begins immediately.
Step 3: Gap Analysis. Which requirements does the company already meet? Many organisations have already built parts of the necessary infrastructure through GDPR compliance – data governance, impact assessments, transparency processes.
Step 4: Implementation Roadmap. Prioritise by risk level and effort required. Tackle high-risk systems with broad user reach first.
Step 5: Governance Structure. Appoint an AI officer or establish an AI board to ensure ongoing compliance – including monitoring, incident reporting, and regular reviews.
The AI Act is widely perceived as a burden. Yet three strategic opportunities are often overlooked:
Export Advantage: AI Act-compliant systems automatically satisfy requirements in most global markets. For European companies selling AI products internationally, compliance becomes a quality seal.
Trust Premium: Customers – especially in B2B – demonstrably prefer verifiably compliant AI solutions. In regulated sectors such as finance, healthcare, and public administration, AI Act compliance is fast becoming a market-entry prerequisite.
Quality Improvement: The AI Act’s documentation and testing requirements compel companies to develop and operate AI systems more systematically. This reduces errors, improves performance, and lowers long-term maintenance costs. Companies treating compliance as a quality initiative gain twice over.
Yes. As a provider, OpenAI must meet the requirements for general-purpose AI. Companies deploying ChatGPT are considered deployers and must fulfil transparency obligations – users must be informed they are interacting with AI. Additional requirements apply if ChatGPT is used in high-risk contexts.
For mid-sized companies with only a few high-risk systems, typical first-year costs range from €50,000 to €200,000. For enterprises operating many high-risk systems – or developing their own AI – the investment can exceed €500,000. SMEs benefit from simplified compliance pathways explicitly provided for in the AI Act.
Each EU Member State appoints national supervisory authorities. In Germany, the Federal Network Agency (Bundesnetzagentur) is expected to take the lead. Additionally, a European AI Office will coordinate enforcement across the bloc. Supervisory authorities may conduct audits, request information, and impose fines.
Yes – if their AI systems are used in the EU or affect EU citizens. The principle mirrors the GDPR: jurisdiction is determined not by the company’s headquarters, but by where the system is deployed. U.S. tech giants must adapt their systems for the EU market.
Existing high-risk systems must comply by no later than August 2027. Systems substantially modified after the AI Act’s entry into force fall under the new rules immediately. Recommendation: Don’t wait until the final deadline – build compliance incrementally.
Header Image Source: Unsplash / Guillaume Périgois
Translated from the German original using artificial intelligence. The German version is authoritative.