What will the EU AI Act do to us? Short answer: surprisingly little

99 percent of quality engineers in automotive aren't really affected by the EU AI Act.

Last week, I got my second sales email. „We'll help you with AI Act compliance, 6-month project, you need to start now, August 2026 is coming."

Before that came the first one, with the same pitch and a different invoice.

So I opened the AI Act. I read the Digital Omnibus. I read what VDA is preparing. I sat with it for half a day and put together the picture nobody is going to charge me thousands of euros for.

For 99 percent of quality engineers in automotive, the answer is surprisingly simple. Don't panic. What you actually need to do by August 2026 is transparency. That's it.

Here's why.

What the EU AI Act is, briefly

The EU AI Act is the European regulation on artificial intelligence. Approved, in force since 1 August 2024. It works on a risk-category principle.

Four tiers.

The first is prohibited practices. Social scoring, manipulation of children, real-time biometric identification in real time. This doesn't apply to quality engineers.

The second is high-risk. This is AI that can affect health, safety or fundamental rights. This is split into two subcategories. Annex III are standalone systems for a specific purpose (HR decision-making, education, law, critical infrastructure). Annex I is AI embedded in regulated products, including automotive. Careful here, we'll come back to it.

The third is transparency. That's Article 50. AI you're communicating with has to tell you it's AI. AI-generated content has to be identifiable. This applies to 99 percent of quality tools.

The fourth is minimal risk. No obligations.

When does what apply, especially after the Digital Omnibus

In May 2026, the European Commission, through the so-called Digital Omnibus, pushed back the dates for the high-risk categories. This matters because before, there was panic that everything had to be done by August 2026.

Here's the current state.

1 August 2024. The AI Act applies.

February 2025. The bans started. Prohibited practices.

August 2025. Obligations started for General Purpose AI models, which are the big foundation models like GPT, Claude, Gemini.

2 August 2026. Transparency obligations under Article 50 begin for end users. This applies to you.

2 December 2026. A small detail worth knowing. For AI systems that were already on the market before 2 August 2026, there's a postponement for one specific obligation, the machine-readable marking of AI-generated content under Article 50(2). Four extra months. For new systems placed on the market after 2 August 2026, everything applies from day one.

2 December 2027. High-risk obligations for Annex III begin. Standalone AI systems for a specific purpose.

2 August 2028. High-risk obligations for Annex I begin. This is the date for automotive.

You can see it. For most quality tools, the relevant date is August 2026 and it's only about transparency. The date for automotive Annex I is all the way to 2028.

Are you high-risk or not?

The simplest test is based on what the AI tool does.

If AI only assists or optimises work, but you're still the one making decisions, you are not high-risk. AI assistant for 8D, PFMEA generator, AI moderator of a 5 Why analysis, AI tool for audit preparation. All transparency tier.

If AI decides on its own about something with safety or health impact, or is embedded into a regulated product, you may be high-risk. AI in a line vision system that stops the line on a defect detection on its own is borderline. AI in ADAS, autonomous driving, braking systems is high-risk under Annex I.

The latest amendment to the safety component definition from 7 May 2026 made it even clearer. The provisional agreement of the European Parliament and the Council of the EU on the Digital Omnibus on AI states that AI systems used solely for user assistance, performance optimisation, efficiency, automation or quality control are not safety components, unless their failure or malfunction could endanger health or safety. That's the Council of EU press release of 7 May 2026, publicly available.

For most quality engineers, that's a relief.

VDA has its own plan

The second thing I was asked at the meeting was VDA. Whether to wait with AI tool implementation until VDA QMC says something on it.

VDA QMC is preparing the so-called yellow volume for AI in Quality Management. The yellow volume is, for VDA, the stage before final approval. The feedback phase is currently running and ends on 30 June 2026. After that, finalisation and publication follow.

The content covers how to place AI in QM into existing structures. PFMEA, 8D, audits, model validation, output oversight.

That's good news. Within half a year, you'll have a VDA framework to refer to during internal and customer audits. No own compliance project in isolation. Just slot it into what VDA proposes.

But you don't have to wait. Transparency and output labelling is good practice even without the VDA paper.

What to do as a quality engineer today

Three steps.

First. Make a list of AI tools your team is using. Whether officially or unofficially. ChatGPT for drafts, AI translator for customer communication, AI in an inspection camera, internal PFMEA generator. Anything that breathes AI. Roughly classify them. An assistant for documentation is transparency tier. AI with a direct safety impact at the machine is borderline Annex III or Annex I.

Second. On the transparency-tier tools, verify disclaimers. The AI Assistant has to tell the reader it's AI. An AI-generated PDF has at least a note in the footer. That's Article 50. Doable today, don't wait for August 2026.

Third. Watch the VDA yellow volume. If you want to influence the content, register for the feedback phase by 30 June 2026. Otherwise, wait for finalisation after the summer and slot it into internal procedures.

And don't ignore the 2028 deadline for Annex I, if you as a supplier are building AI components embedded into a vehicle. There, you need to start preparing the technical documentation now, because the type-approval audit won't come overnight.

What we did in QualityOS

And now the practical question that comes to me most often. So how does your QualityOS handle it?

QualityOS is a decision-support assistant. It helps you structure an 8D, analyse root cause, prepare an audit response. But the final decision is still yours.

In the category under the AI Act, we're in transparency tier. We meet it today, we're not waiting for August 2026.

There's a disclaimer under every AI answer. „QOS is AI, the final decisions are always yours." Every AI-generated PDF is labelled. In preparation is a VDA yellow volume module, which will be loaded the moment VDA QMC publishes the final version.

And when your customer comes for an audit and asks „do you have AI in the process?", the answer isn't defence. The answer is „yes, we do, it's decision-support tier under the AI Act, transparency met, oversight kept."

That's exactly what the auditor wants to hear. Yes, we use it. Yes, we know what category it falls in. Yes, we maintain transparency.

Conclusion

The EU AI Act isn't a reason to panic. It isn't „a massive compliance project for 2026" either. For most quality engineers in automotive, it's about transparency. A disclaimer, output labelling, clarity on who carries the final responsibility. That's all by August 2026.

The VDA yellow volume will arrive within half a year. It will be useful. But you don't have to wait for it to do the basics.

And if you build AI components embedded directly into a vehicle, you have the 2 August 2028 deadline for high-risk obligations under Annex I. That's no longer about transparency. That's technical documentation, model validation, oversight. Start this year, not in 2028.

For everyone else, peace. Your 8D assistant won't make a regulatory nightmare for you. Just let it announce it's AI.

And that's the whole August 2026.

---

The information is a paraphrase of the state as of June 2026. Sources: EU AI Act (Reg. 2024/1689), Digital Omnibus on AI (provisional agreement, 7 May 2026, Council of EU press release on consilium.europa.eu), Regulation 2018/858, Regulation 2019/2144, draft European Commission guidelines on AI transparency under Article 50, VDA QMC publication plan (yellow volume AI in QM, feedback phase to 30.6.2026). This text is not legal advice. For specific assessment of the risk category of your AI, consult a compliance specialist or notified body.