Community Smart Hub · AI & You Series
Published 21 June 2026 · 5 min read · No tech background needed
Imagine watching a glowing product review from a smiling shopper — only to discover that person was never real. They were generated entirely by AI. This week’s news covers exactly that, plus important new laws protecting people from harmful deepfakes.
1. Brands Are Using Fake AI “People” to Advertise Products
The Guardian newspaper revealed this week that some companies are paying to create entirely artificial people — generated by AI — who then appear in social media videos pretending to be real customers. They give five-star reviews. They share “genuine” unboxing videos. They appear to be brides at weddings, ordinary shoppers, or trusted friends recommending a product. None of them exist.
AI-generated influencer — A computer-created person who looks and sounds real but has never actually lived, breathed, or used the product they are promoting.
The legal gap: Adverts in the UK must not mislead people — but the UK Advertising Standards Authority (the body that regulates adverts) has confirmed there is no rule requiring companies to say an advert was made with AI. That gap is exactly what some brands are quietly exploiting.
How a Fake AI Review Makes Its Way to Your FeedBrand usesAI to generatea fake person”Person” recordsglowing review(NDA signed)Posted on socialmedia — lookscompletely realYou see it.No label. Nowarning.
Why this matters to you: Fake AI reviews can push you towards products that don’t work, or services that don’t deliver. They exploit trust — and they are currently very hard to spot without specialist tools.
2. Europe Bans Harmful Deepfakes and Tightens Rules
On 16 June 2026, the European Parliament voted — 423 votes in favour, 57 against — to pass new protections into law. Here is what has changed:
Banned outright: AI systems that can generate child sexual abuse material, or non-consensual sexual images, video or audio of real identifiable people (often called “nudifiers” or “revenge deepfakes”). Companies, not just individual bad actors, must now put proper safeguards in place.
Deepfake — A video, audio clip, or image that has been manipulated by AI to make it look or sound as though a real person said or did something they never actually said or did.
Non-consensual sexual image — A sexual image of a real person created or shared without their knowledge or permission. Creating these with AI is now banned in Europe.
Deadline update: The rule requiring AI-generated content to carry a machine-readable label has been pushed back from August 2026 to 2 December 2026. This gives companies more time to prepare — but the clock is still running.
European Deepfake Law: What Is Now Banned, What Is Coming SoonNOWJune 2026Nudifier banpassed into lawSOON2 Aug 2026AI transparencyrules take effectCOMING2 Dec 2026Machine-readablelabels requiredon AI content
3. Why Spotting a Fake Voice Is Harder Than It Sounds
Researchers at a major AI conference found that the way we currently try to identify which AI tool created a fake voice is flawed. The current method — comparing voices in pairs, like a police line-up — turns out to be less accurate than a completely different approach that looks at all known voices at once.
Pairwise comparison (biometric matching) — A method of identifying something by comparing two examples side by side: “Is Sample A more like Source 1, or Source 2?” Used widely in fingerprinting and voice recognition.
The research found that grouping all known AI voice generators together and looking for patterns works significantly better — reducing errors from roughly 12–15% down to 8.6%.
4. A New Test Set for Fake Documents — Not Just Fake Faces
Most deepfake research focuses on fake videos of faces. But a new research collection — the CIFAR Synthetic Evidence Corpus — focuses on something different: fake documents. Think altered receipts, fabricated messages, or tampered official records. These can change the outcome of legal cases or fraud investigations.
Real-world implication: A tiny change to one field in an official document — a date, a sum of money, a name — can have enormous consequences. Current deepfake detectors are not trained to spot this. This new dataset starts to fix that.
5. Using AI to Check AI Can Make Things Worse
Researchers at an AI conference found a worrying pattern: when companies use an AI system to “quality check” the data that another AI is learning from, it can accidentally make the second AI worse — not better.
The problem: the quality-checker only looks at a limited range of examples. It approves content that looks similar to what it already knows, and rejects rare or unusual examples — the very examples the AI needs to understand the world properly.
Tail modes — Rare but important examples in a dataset. Just as most people are average height but some are very short or very tall, most data points are common but a few are rare. Losing rare examples makes an AI less able to handle unusual situations.
📌 This Week’s Theme in Plain English
This week, AI is being used to deceive — fake reviews, fake people, fake voices, fake documents. But laws are catching up. The European Union is drawing firm lines. And researchers are finding better ways to detect fakes, attribute content to its true source, and protect people from harm.
Community Smart Hub exists to make sure your community knows what is happening — and what you can do about it.Community Smart Hub · AI & You Series · Blog 2 of 5 · Week of 20 June 2026
Sources: The Guardian, European Parliament, Interspeech 2026, CIFAR, ICML 2026
