Community Smart Hub · AI & You Series
Published 24 June 2026 · 5 min read · No tech background needed
The biggest change happening in the world of AI right now is not a new chatbot or a faster computer. It is a fundamental shift in the question we ask about media. We used to ask: “Is this real or fake?” The new question is: “Who made this, and can we prove it?” This blog explains why that shift matters — and what it means for you.
1. We Are Moving Beyond the Simple “Is It Fake?” Test
For years, researchers have tried to build a single detector that could look at a photo or video and say: “Real” or “Fake.” This week, a new benchmark — a standard test called OpenFake — was released that shows just how difficult this has become.
Benchmark — A standard set of tests used to measure how good a detector is. Like a driving test: it checks whether a system can handle a set of known challenges. But just like a driving test does not guarantee you can drive in every city in the world, a benchmark cannot cover every possible fake.
The OpenFake benchmark uses politically themed deepfakes created by the most advanced modern AI tools — and a human study found that many of the fake images were effectively indistinguishable from real ones, even to trained observers.
What this tells us: Future deepfake detectors need to be tested against content they have never seen before — not just variations of content from the same tools they were trained on.
2. The New Standard: Five Questions Instead of One
Rather than asking one big question (“Is this fake?”), experts now say we should ask five smaller, specific questions about any piece of media. If we can answer all five, we have genuine evidence of trustworthiness.
The Five Questions of AI Media TrustMEDIAunder review① Who created this?Can we name the sourceand verify their identity?② Is provenance verified?Is there a cryptographiccertificate we can check?③ Watermark intact?Does the hidden markerstill verify correctly?④ Logs match?Do generation recordsconfirm the source tool?⑤ Detectors agree?Do independent forensictools reach same result?
Plain English: Imagine buying a second-hand car. You do not just look at it and guess. You check the logbook, run a history check, look at the VIN number, get a mechanic to inspect it, and check the insurance database. Five checks, not one.
3. Watermarks Are Getting Smarter — They Now Identify the Creator
The SAiW watermarking system — short for Source-Attributable Invisible Watermarking — has gained significant attention this week. Unlike a standard watermark that merely says “AI made this,” SAiW can answer a much more specific question: “Which AI system made this, and when?”
Source attribution — The ability to trace a piece of content back to the specific tool or system that created it. Like a hallmark on jewellery that names the maker, not just the material.
Crucially, the watermark is designed to survive the kinds of changes content routinely undergoes online — compression, cropping, colour adjustments, and even deliberate attempts to remove it.
4. The UK Government Identifies Who Needs Deepfake Detection Most
The UK government published its latest technology overview this week, identifying the areas where deepfake detection is in highest demand right now:
🏦
Fraud prevention — Banks and lenders using AI to detect fake identity documents or synthetic identities.
🔐
Identity verification — Confirming that a person in a video call or a selfie is who they claim to be.
📱
Content moderation — Social media platforms scanning for harmful synthetic content before it spreads.
🛡️
National security — Intelligence services tracking AI-generated disinformation campaigns.
The broader point: Deepfake detection is no longer a niche research topic. It is becoming essential infrastructure — like a firewall for a computer, or a passport check at a border.
5. Protect at Creation, Not Just Detection After the Fact
A paper published in Scientific Reports this week proposed what researchers call “proactive” protection of videos: rather than trying to detect fakes after they have spread, protect the video at the exact moment it is recorded.
The system uses three components working together:
Proactive Video Protection: How It Works🧠 AttentionmechanismSpots the mostimportant partsfeeds into🔏 InvisiblewatermarkHidden inside thevideo at creationverified by🔗 Blockchain recordPermanent tamper-proof logthat cannot be alteredeven by the creator
📌 The Big Picture: A Week in Summary
This week’s five briefings point to a single, clear direction: the world is shifting from asking “Is this fake?” to asking “Who is responsible for this, and can we prove it?”
That shift changes everything — from how cameras are built, to how adverts are labelled, to how courts consider evidence, to how your identity is verified online.
Community Smart Hub believes that understanding these changes is not just for tech experts. It is for everyone — parents, shoppers, voters, young people, and pensioners alike. Because the decisions being made right now, in research labs and parliament buildings, will shape what you see online for the next decade.
Thank you for reading this week’s series. Share it with someone who needs to know.Community Smart Hub · AI & You Series · Blog 5 of 5 · Week of 20 June 2026
Sources: OpenFake benchmark, SAiW research, UK Government technology overview, Scientific Reports
