The Invisible AI Map
Think of the photos of you that have drifted online over the years: a team picture, a friend’s public wedding album. Now picture a company you have never heard of copying them, turning your face into a biometric code and adding it to a face-search database. Nobody asked you. Nobody told you.
That scene is illustrative. The practice is not. In September 2024, the Dutch data protection authority fined the US company Clearview AI €30.5 million. According to TechCrunch, the regulator found that Clearview had scraped about 30 billion images from the internet, turned faces into biometric codes without a valid legal basis, and failed to tell people or answer their requests to see their data. The authority’s chair warned that “this makes using the services of Clearview illegal” for Dutch organizations. Clearview said it was not subject to European data protection law and called the decision unenforceable (Lomas, 2024).
Here is my proposition; test it against the documents below. Europe’s AI law is not a future law. It already protects your face. But the protection that matters most for your job, your loan and your access to public services is still a promise with a date on it: December 2027.
The ban that already bites
The Clearview fine was issued under the EU’s data protection law, the GDPR. Since 2 February 2025, a second law speaks to the same practice much more bluntly. The EU AI Act now simply bans AI systems that create or expand facial recognition databases through “the untargeted scraping of facial images from the internet or CCTV footage” (Regulation (EU) 2024/1689, art. 5(1)(e)). Breaking a ban like that can cost up to €35 million or 7 percent of a company’s worldwide annual turnover, whichever is higher (art. 99(3)).
If you have heard that the AI Act “has not started yet”, that is the first correction. Parts of it have applied for well over a year.
How the law decides what to worry about
The AI Act sorts AI uses by how much harm they could do. The European Commission describes four levels (European Commission, n.d.):
- Unacceptable risk: a short list of practices considered a clear threat to people’s safety, livelihoods and rights. These are banned.
- High risk: uses that can seriously affect health, safety or fundamental rights, such as AI used in hiring, education, access to essential services or law enforcement. These face strict obligations.
- Transparency risk: systems where people should be told what they are dealing with, such as chatbots and deepfakes.
- Minimal or no risk: most everyday AI, such as spam filters or AI in video games. No specific new rules.
Separate rules cover the large “general-purpose” models that power chatbots and many other tools.
Key idea: “in force” versus “applies”
A law enters into force when it officially becomes law. Its duties apply from the dates written into it, which can be months or years later. The AI Act entered into force on 1 August 2024, but its rules switch on in stages. For you, that means the right question is never just “is there a law?” but “does this particular rule apply yet, and to whom?”
The receipts: what switched on, and when
The dates below reflect the Act as amended this summer by a package known as the AI Omnibus (Regulation (EU) 2026/1744), which entered into force on 27 July 2026 and moved several deadlines (Cuatrecasas, 2026; European Commission, n.d.).
1 August 2024: the law enters into force. Nothing changes for you yet, but the clock starts.
2 February 2025: the bans begin. Prohibited practices include manipulative techniques that cause significant harm, exploiting people’s age, disability or economic situation, social scoring, predicting someone will commit a crime based solely on profiling or personality traits, untargeted scraping for face databases, emotion recognition in workplaces and schools (except for medical or safety reasons), inferring sensitive traits such as religion or sexual orientation from biometric data, and, with narrow exceptions, real-time remote biometric identification by police in public spaces (art. 5(1)). An AI literacy duty also began. Since the Omnibus, it requires organizations to take measures supporting their staff’s AI literacy, without prescribing a particular level (Cuatrecasas, 2026).
2 August 2025: rules for general-purpose AI models. Providers of large models took on documentation and transparency duties. Models already on the market before that date have until 2 August 2027 to comply (art. 111(3)).
2 August 2026: transparency rules. Chatbots must be designed so people know they are talking to an AI unless that is obvious. Organizations using deepfakes must disclose that content was artificially generated or manipulated. Providers of systems that generate synthetic images, audio, video or text must mark the output in a machine-readable way (art. 50). Generative systems already on the market before August 2026 have until 2 December 2026 to add that marking (art. 111(4)).
2 December 2026: a new ban. The Omnibus added a prohibition on AI used to generate non-consensual sexually explicit images of real people and child sexual abuse material (Abreu Advogados, 2026; European Commission, n.d.).
2 December 2027: high-risk rules for stand-alone systems. This is the one that reaches ordinary lives. AI used in areas such as hiring, education, credit and access to public services must meet requirements on risk management, data, records, human oversight and accuracy. Organizations using these systems must assign competent human oversight, monitor them, keep logs and, for certain decisions, tell the people affected (art. 26). Affected people gain a right to a clear and meaningful explanation of the AI system’s role in a decision about them (art. 86).
2 August 2028: AI inside regulated products. High-risk AI built into products already covered by EU safety law, such as machinery or medical devices, follows on this date.
Older systems. High-risk systems already on the market before 2 August 2026 generally come into scope only if their design changes significantly, but those used by public authorities must comply by 2 August 2030 (art. 111(2)).
Who gets the wait, and who carries it
Read that timeline as a person, not a lawyer. The rules about what you can see (chatbot labels, deepfake disclosure) came first. The rules about what is decided about you (the job, the loan, the public service) come last. Until December 2027, the organizations running those systems get time to prepare. The people on the receiving end of their decisions get to wait for a right to an explanation. That is my reading, not a finding.
The best argument against me
Here is the strongest case for the slow rollout. The high-risk rules demand real work: risk management, data controls, records, human oversight and accuracy. Much of the Act will be shaped by guidance, standards and national authorities that are still being finalised. Rush it, and you may get paperwork instead of protection.
Fair point. But it cuts both ways. The Omnibus has already shown that dates can move. A runway that keeps getting longer stops being a runway.
The verdict
On the evidence, the AI Act is live. The bans apply now, not in 2027. Today, in the EU, you should be told when you are chatting with an AI rather than a person, and deepfakes should be labelled. From December 2027, if an AI system on the high-risk list helps decide whether you get a job, a loan or a public service, you can ask the organization that used it to explain the system’s role. None of that happens automatically: rights work when people use them.
For organizations, December 2027 is a runway, not a reprieve. The first hard question is classification: is a given tool really just a helper, or does it shape decisions about people? Chapter 3 of Operational AI Readiness warns that “a workflow does not become low consequence just because the team gives it a soft name” (Bakos, 2026). The second is evidence. Chapter 11 describes what reviewers tend to ask, and it is rarely abstract: who owns this, what changed, where is the human review, and finally, “Show me the record” (Bakos, 2026). Records cannot be created retroactively.
What we still don’t know
The Clearview fine was issued under the GDPR before the AI Act’s bans applied; it shows the kind of practice the ban targets, not an AI Act enforcement action. We did not find, and do not report, any penalty yet imposed under the AI Act itself. This guide is a plain-language summary, not legal advice; for any decision that matters, check the official text.
Which leaves the question we are watching: when the first penalty under the AI Act itself lands, who will it be, and will it be about something that touches your face, your feed or your paycheck?
Questions you can ask
- If you are in the EU: Am I talking to a person or an AI? If you are not sure, ask; the answer should be clear.
- If a decision affects you: Was an AI system used, and is it one the AI Act treats as high-risk?
- If your organization uses AI: Have we checked every system against the banned list, which already applies?
- If your organization uses AI: Which of our systems would be high-risk in December 2027, and what records are we keeping now?
Go further. Chapter 3 of Operational AI Readiness shows how to classify decision-influencing AI honestly, and Chapter 11 sets out what auditors, regulators and internal reviewers will actually ask to see. Read the book free, or test one workflow with the free readiness assessment.
Stay ahead of the dates. Guidance, standards and enforcement decisions under the AI Act will keep arriving between now and 2028. Mercury Intelligence ($10 a month, or $100 a year) gives you our full investigations and analysis as they land. See what it includes.
FairByDesign is reader-funded. If this helped, get the free Mercury Brief or see ways to support. Support never buys editorial control.
References
Abreu Advogados. (2026). Digital Omnibus Regulation on AI: The key amendments to the AI Act. https://abreuadvogados.com/en/conhecimento/publications/digital-omnibus-regulation-on-ai-the-key-amendments-to-the-ai-act/
Bakos, V. (2026). Operational AI readiness: A practitioner guide to governing AI that is already live. FairByDesign. https://fairbydesign.org/operational-ai-readiness/
Cuatrecasas. (2026, July 24). Digital Omnibus on AI has been published. https://www.cuatrecasas.com/en/spain/intellectual-property/art/digital-omnibus-ai-has-been-published
European Commission. (n.d.). AI Act [Regulatory framework for AI]. Shaping Europe’s digital future. Retrieved October 15, 2026, from https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
Lomas, N. (2024, September 3). Clearview AI hit with its largest GDPR fine yet as Dutch regulator considers holding execs personally liable. TechCrunch. https://techcrunch.com/2024/09/03/clearview-ai-hit-with-its-largest-gdpr-fine-yet-as-dutch-regulator-considers-holding-execs-personally-liable
Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), as amended by Regulation (EU) 2026/1744. Official Journal of the European Union. http://data.europa.eu/eli/reg/2024/1689/oj (articles 5, 26, 50, 86, 99 and 111 consulted via https://artificialintelligenceact.eu/)




Leave a Reply
You must be logged in to post a comment.