Two developments in Europe this week capture a tension that will define AI governance for years to come.
On 12 August, German digital-rights organisation HateAid filed a criminal complaint over Meta’s Ray-Ban Meta Wayfarer Gen 2 smart glasses, arguing they enable covert recording in violation of German privacy law. The complaint was submitted to ZIT, Frankfurt’s specialist cybercrime prosecution unit, and names Meta management, EssilorLuxottica units including Ray-Ban and Oakley, and four retailers: Fielmann, Apollo-Optik, Mister Spex and MediaMarkt. At almost the same time, Anthropic began embedding invisible statistical watermarks into text generated by new Claude models worldwide, complying with EU AI Act Article 50 transparency requirements that came into force on 2 August. One dispute concerns AI that can observe people without their knowledge; the other concerns AI-produced work that may pass unnoticed as human. Together they point towards the same underlying problem: if AI is going to become ambient, Europe increasingly wants it to remain detectable.
The technology becomes more useful as it becomes less intrusive. A pair of glasses can record, interpret and respond to what its wearer sees whilst appearing indistinguishable from ordinary sunglasses. A chatbot can produce an essay, report or email that may be almost impossible to distinguish from human writing. In both cases, seamlessness is the product — but seamlessness also removes the social signals that normally allow people to know when they are being observed or when they are reading machine-generated content.
Smart Glasses Change the Social Contract
HateAid’s legal basis is specific. Germany’s Telecommunications, Digital Services and Data Protection Act (TDDDG) bans the placing on the market of communication devices disguised as everyday objects and designed to film people without their knowledge.
Violations carry a fine or up to two years in prison under Section 27. HateAid argues that the Ray-Ban Meta Wayfarer Gen 2, built on one of the best-selling sunglasses ever made, is close to impossible to distinguish from the camera-free version. HateAid managing director Josephine Ballon put it in direct terms: “There’s no place to escape from smart glasses. You have to expect at any moment to be filmed and then exposed on the internet.” Meta told Politico EU that its glasses are “built with privacy safeguards from the ground up that go beyond what any smartphone offers.” ZIT confirmed receipt and said it would conduct a preliminary assessment of whether grounds exist for a deeper investigation.
The Federal Network Agency’s position is that smart glasses are legal provided recording is unmistakably signalled, typically through a visible light. Hamburg’s data protection commissioner Thomas Fuchs tested the glasses and found the recording LED insufficiently visible in everyday use, saying in late July that a sales ban was possible. France’s CNIL went further, issuing a public warning on 11 May 2026 that smart glasses posed a “significant risk” of normalising surveillance that is “almost invisible and omnipresent,” launching a dedicated action plan. The timing also matters for a wider market question: Samsung’s Galaxy Glasses are announced for autumn 2026, using an external design visually indistinguishable from standard eyewear.
If Germany’s legal standard requires that bystanders be able to reliably identify a camera, it would apply across the entire category.
AI Writing Creates the Opposite Problem
With generated text, the invisibility works in reverse. Instead of AI observing a human unnoticed, a human reader may unknowingly consume work produced by AI.
Anthropic has responded to EU AI Act Article 50(2) obligations by embedding an imperceptible statistical watermark directly into text generated by Claude models launched on or after 2 August 2026. The mark travels with text when it is copied and pasted and may survive some editing. Supported file formats such as .svg, .png and .jpg can also carry signed C2PA provenance metadata. The system applies globally, covering Claude’s consumer chatbot, the developer-facing API, Claude Code, Claude Cowork, Claude Tag and Claude models accessed through AWS, Google Cloud and Microsoft Foundry. A text detection API allowing third parties to check for the mark is coming. Non-compliance with Article 50 can trigger fines of up to €15 million or 3% of worldwide annual turnover.
Anthropic is the first major frontier AI lab to deploy production-scale text watermarking across all its products simultaneously. OpenAI’s own EU compliance statement, published this month, confirmed that text watermarking remains harder to deploy at scale and had not yet been rolled out. That gap reflects a genuine technical challenge: text gets copied, paraphrased, translated and folded into other writing in ways that images do not. Anthropic itself acknowledges the limitations: heavy editing, translation, format conversion and screenshots may all remove the signal. Older Claude models not yet covered.
And the Brussels effect is visible here: Anthropic is applying the system globally because maintaining separate versions for different markets is impractical, extending European standards to every Claude user worldwide.
Detection Is Not Proof
Watermarking sounds like a clean answer, but it is not.
Anthropic itself warns that a detected mark does not prove Claude wrote an entire document: asking Claude to proofread or translate a paragraph could leave a trace on fundamentally human work. The reverse also applies: heavy editing may weaken or remove the signal. For schools and workplaces tempted to treat detection tools as evidence of academic or professional dishonesty, this creates a significant liability. AI detectors have already produced false positives, particularly when analysing formulaic or non-native English writing. A watermark originating from the model provider is potentially more reliable than a third-party detector, but an Anthropic engineer confirmed on 12 August that the mark “is not perfect, you can edit it, but it’s a first step.”
Detection therefore needs context. A student asking AI to write an entire assignment and submitting it unchanged is different from using the same system to correct grammar. A journalist generating a fictitious interview is different from using AI to transcribe a real one. Regulation cannot replace those distinctions with a binary label. Regulators expect providers to rely on a multi-layered approach combining invisible watermarks, signed metadata and, where possible, fingerprinting or logging, because no single technique is sufficient on its own.
Europe is Building a Right to Know
What connects smart glasses and watermarking is the principle of informed participation.
If someone is being recorded, they should have a reasonable chance to know it. If a reader is consuming AI-generated material, there should be some mechanism for revealing its origin. That does not require every AI-assisted sentence to carry a warning label or every pair of smart glasses to become socially unacceptable. AI can improve accessibility, translation, memory, navigation, creative work and productivity. Smart glasses could be especially valuable for people with visual impairments or workers who need hands-free information. But seamlessness cannot become an excuse for secrecy.
The central European instinct emerging from these cases is that AI may become invisible in operation without becoming invisible in accountability. The technology can disappear into glasses, documents and software, but its presence should remain discoverable when it affects other people’s privacy, judgement or trust. That principle will become harder to enforce as AI improves: cameras will become smaller, generated writing more natural and detection easier to evade. What Europe can establish is an expectation: when artificial intelligence crosses from a private tool into somebody else’s privacy or trust, people should have a way of knowing it is there. The challenge is no longer simply making AI intelligent. It is making increasingly invisible intelligence socially legible.
Keep up with Daily Euro Times for more updates
Read also:
Bubble Swells: Europe Insists On Labelling AI Content
Could AI Follow the Metaverse Into Oblivion?
Big Tech Giants Take Over the European Public Square
