New legal requirements are now in force on both sides of the Atlantic, obliging AI companies and the businesses that use their tools to label content produced or substantially edited by AI systems. The rules also require AI providers to offer public tools so that anyone can check whether a piece of content was machine-made. Compliance has been uneven. And the technical method at the centre of one proposed labelling approach has already drawn serious criticism. This is not a story about AI capability. It is a story about whether the information reaching you carries any honest signal about where it came from.

What happened

Two separate frameworks are now active. One operates at EU level and covers AI systems broadly. The other is a California law focused specifically on transparency about AI-generated content. Together, they place legal obligations on AI providers and on companies that deploy AI tools to label output that was created or substantially shaped by those systems.

Part of the requirement is that AI companies make detection tools publicly available — so that an ordinary reader, not just a regulator, can check a piece of content. According to one monitoring report, only some companies are currently meeting that obligation.

One proposed labelling method attracted particular criticism: embedding hidden patterns into AI-generated text to mark it as machine-made. A prominent technology commentator argued that this approach would corrupt ordinary writing and could misfire, flagging text written by humans as AI output. That is not a minor technical quibble. If the detection layer is unreliable, the entire labelling system loses its credibility.

Separately, a major AI chat and search tool was formally classified as a very large online search engine under EU digital rules. That classification matters because it reflects how central the tool has become to how people find and first encounter information.

Who is affected

In practical terms, this affects anyone who reads anything online — news, opinion, social media posts, search results. The rules exist to tell you when what you are reading was made by a machine. If they work, you know. If they do not, you do not.

AI companies face direct legal obligations and reputational exposure if they are found non-compliant. Enforcement bodies now have to decide what to do about the companies that are not yet meeting the standard.

There is a secondary risk that falls on journalists, writers, and anyone who produces text professionally. If detection tools misfire — which the criticism above suggests is a real possibility — their human-written work could be wrongly labelled as AI output. That kind of false flag is not just an inconvenience. In a dispute about credibility, it could be used as a weapon.

What the real risk is

The core promise of these laws is simple: readers will know when content is machine-made. If companies do not comply, or if their detection tools are unreliable, that promise is empty. Readers stay in the dark. The label that was supposed to help them becomes noise.

Faulty detection creates a second, sharper problem. A tool that mislabels human writing as AI output can be used to discredit genuine voices. That tactic — casting doubt on a real person’s work by suggesting it was manufactured — is exactly the kind of move that AI-generated misinformation ecosystems make possible at scale.

It is also worth being clear about what a label does not do. A clearly labelled AI article can still push a false narrative. The label is a starting point, not a verdict on accuracy.

The classification of a major AI tool as a search engine signals that regulators see it as shaping what information people encounter first. That raises the stakes for how its outputs are framed — and whether those outputs carry any warning at all.

What to do today

These are concrete steps you can take this week.

• Read for texture. Ask yourself whether a piece reads as though someone with specific knowledge and experience wrote it, or whether it feels generic and frictionless. Machine output tends toward the latter. That is not proof, but it is a signal worth noticing.
• Use the detection tools that exist. If an AI company is legally required to offer a public detection tool, try it on content you are uncertain about. Treat the result as one data point, not a final answer, given the reliability concerns raised above.
• Check the claim in older sources. Search for the same underlying fact in archived news articles or academic sources that predate the current AI content boom. If the claim only appears in recent, undated material, that is worth noting.
• Notice the emotional pressure. Pay attention to whether a piece is pushing you to feel something urgently — fear, outrage, certainty — before it gives you any evidence. That pressure is a persuasion tactic. It works whether the author is human or machine.
• Follow enforcement coverage. Monitoring reports and coverage of regulatory actions, including the role of online intermediaries in disinformation contexts, will tell you how these rules are actually being applied rather than how they are written.

Why this keeps happening

Regulation arrives after technology has already been deployed at scale. That is not an accident or an oversight — it reflects how long it takes democratic institutions to move compared to how quickly a product can be shipped. The rules are always catching up.

Companies have a financial incentive to keep content pipelines moving. Building and maintaining public detection tools adds cost and friction. Some companies are slow to absorb that cost. The monitoring report showing uneven compliance is the predictable result.

The technical difficulty is also genuine. Distinguishing machine-made text from human-written text is an unsolved problem. Any law that depends on reliable detection inherits that uncertainty. The criticism of hidden-pattern labelling is a direct example of this: a method that sounds precise turns out to be fragile.

The deeper pattern is older than AI. Disinformation has always exploited the gap between what a rule says and what actually gets enforced. The same structure appeared in election interference cases and in coordinated inauthentic behaviour campaigns: the rule exists, the violation is visible, and enforcement lags. AI-generated misinformation is the latest content to move through that gap.

The gap persists partly because online systems were not built with any reliable way to confirm who is actually speaking. A message with no accountable source behind it carries the same visual weight as one with a real name and a verifiable record. Until that changes, the label on a piece of content is only as trustworthy as the system issuing it.

Frequently asked questions

If AI content is labelled, does that mean it is automatically untrustworthy?

No. A label tells you about the origin of content, not its accuracy. A labelled AI article can be factually sound. An unlabelled human-written article can be full of errors or deliberate distortions. The label is useful context, but it is not a quality score.

Can AI detection tools reliably tell me whether something was written by a machine?

Not with certainty. Distinguishing machine-made text from human-written text is technically difficult, and the criticism of hidden-pattern labelling methods suggests that current approaches can misfire. Use detection tools as one input among several, not as a definitive answer.

Why are some AI companies not yet offering the public detection tools the law requires?

According to the monitoring report referenced in coverage of these rules, compliance has been uneven. The reasons have not been fully disclosed publicly, but building and maintaining such tools adds cost and operational complexity. Enforcement bodies now face the question of what happens next for those that have not yet complied.

Originally reported by disinfo.eu. This article summarises that reporting and adds practical guidance.

Scams, fraud, bots and manufactured noise keep spreading because the internet was built with no reliable way to know who anyone actually is. Everyone deserves authenticity and accountability online, and that is the mission we are working on. Subscribe to stay across how these rules develop and what enforcement actually looks like in practice.