SynthID Caught a Sri Lanka Page Faking Australian News
A Sri Lankan Page Is Faking Australian News — Google's Own Detector Busted It
By Daniel Rao | TechPulse USA | 13 October 2026
It called itself "your go-to source for breaking news, politics, sports, and weather across Australia." It posted several times a day, always riding whatever Australia was already talking about — a crash, a stabbing, a soldier's death, a theatre fire. There was just one problem: the page was being run from Sri Lanka, and its "photos" of breaking news were AI-generated fakes.
Now Google's own AI detector has fingered every single one of them.
The "news outlet" reporting from 8,000 kilometres away
The page is called "Aussie News Today," and on the surface it looks the part. An Australian-sounding name, a bio promising round-the-clock breaking news, a posting rhythm of multiple updates a day about genuinely high-profile events. If you scrolled past it in your feed, you would have no reason to suspect a thing.
That's precisely what makes this format so rotten. The page didn't invent the news — the car crash, the stabbing, the training death, the theatre fire all really happened. It invented the evidence: dramatic images purporting to show the scenes, every one of them fabricated. Readers got real headlines stapled to fake photography, and the emotional punch of a disaster photo doing what disaster photos do — making you feel like you were there.
So where is "there," exactly? Not Australia. Meta's page-transparency panel — the bit of a Facebook page that publicly lists the countries where the people managing it are based — shows "Aussie News Today" is operated by a user in Sri Lanka. The bio said Australia. The servers of truth said otherwise.
AAP FactCheck, an accredited member of the International Fact-Checking Network, published its debunk of the page today, and it is a methodical piece of work: four posts, four fakes, each one checked against real footage, real maps, and — in a delicious twist — Google's own AI-detection tool.
Four fakes in fourteen days
October 3 — the Newcastle crash. A car ploughed into a crowd gathered for the Newcastle Knights rugby league grand final procession. "BREAKING: Car Crashes Into Newcastle Knights Fans in Newcastle — 9 Injured," the page announced, alongside a photo of chaos: emergency workers, a police cordon, the works. The crash was real. The photo wasn't. It bore no resemblance to the genuine photos and video ABC News published from the scene that morning, and it carried the classic AI tell — garbled writing on the emergency workers' uniforms and on the police cordon. The model can render a paramedic. It can't spell his jacket.
October 5 — the Newtown stabbing. Emergency services were called to the Dendy Newton cinema in Sydney's Newtown after a 26-year-old woman allegedly stabbed a 48-year-old man in the hand, according to 9News. The page's image showed a man being loaded into an ambulance outside a cinema on a street corner with the wrong logo. The real Dendy Newton isn't on a street corner, and its logo looks nothing like the one in the fake. If you've ever queued for a film in Newtown, the geography is wrong before you even notice the branding — which tells you something about how little care went into the lie.
October 8 — a soldier's death. This is the one that should make you angry. Corporal Brendan Hannam was killed on October 5 when a vehicle rolled over near Darwin during a military training exercise; five others were injured. Three days later, the page posted an image purporting to show the aftermath — paramedics, crash investigators, a military helicopter — with distorted writing smeared across the vehicles. No credible news outlet ever published such an image, because it never existed. A family's tragedy, repackaged as content by a page on the other side of the planet.
September 26 — the theatre fire. The page even posted an AI-generated video report claiming to show a fire at the abandoned Austral Picture Theatre in Collingwood, Melbourne. The footage shows a grand building ablaze. The real theatre is a dilapidated shell that looks nothing like it — you can see as much on Google Maps. A fire, a landmark, a video: all synthetic.
Caught by the company whose tools made the fakes
Here's the part of the story that deserves a slow read. AAP ran the page's images through Google's SynthID Detector — the free tool Google DeepMind opened to the public worldwide on October 7 at synthid.com — and it confirmed that all of them were created or edited using Google AI tools.
SynthID works by embedding an invisible watermark into AI-generated media at the moment of creation — a statistical signal spread through the pixels, not a logo stamped on top — which the detector can then read back. Google says it has watermarked more than 180 billion images and videos since introducing the technology in 2023, and that the detector also reads watermarks from partner companies OpenAI, NVIDIA, and Kakao, with Apple support described as coming soon. The same verification already runs inside Google Search, the Gemini app, and Chrome, handling more than a million checks a day.
"With the new SynthID Detector, anyone can easily check if an image, video, or audio file was made with AI from Google or our partners, including OpenAI, NVIDIA, Kakao, and soon, Apple," Pushmeet Kohli, VP of Science and Strategic Initiatives at Google DeepMind, said in announcing the launch.
Two honest caveats before anyone gets carried away. First, the detector is not a general-purpose AI lie detector — it only finds the SynthID watermark, so a "not detected" result proves nothing; content from systems that don't watermark sails straight through. Second, you need to sign in, and there's a daily usage quota. But in this case the result was unambiguous, and faintly poetic: fakes generated with Google's tools, exposed by Google's detector, six days after the detector went public.
This is not a one-off. It is an industry.
If you think one Sri Lankan page faking Australian news is a curiosity, Le Monde would like a word. In a September investigation, the French paper reported on a sprawling, profit-driven business of AI-generated fake-news pages on Facebook — and Sri Lanka keeps showing up as the factory floor.
The numbers are staggering. The NGO Reset Tech published a study on September 9 identifying more than 1,000 French-language pages created by a single network of individuals living in Sri Lanka; Le Monde's own estimate put the total at more than 2,000 pages targeting France. One Sri Lankan network aimed at France alone comprised 1,340 pages that had accumulated 36 million reactions, 6.2 million comments, and 1.9 million shares. Smaller operations, Le Monde found, run out of Venezuela, Vietnam, Algeria, and Pakistan. The motive isn't politics — it's money. Divisive, emotional content earns reactions, reactions earn reach, and reach earns ad revenue or traffic to ad-stuffed "news" sites.
Le Monde even named a central figure: Geeth Sooriyapura, a Sri Lankan entrepreneur and influencer who claims to have made more than $300,000 from the fake-page business and now trains a new generation of page creators through workshops and coaching sessions. Asked about his role, he denied any involvement and claimed identity theft — though Le Monde viewed training videos in which he could be seen personally running several pages aimed at French audiences.
And Meta? The platforms, Le Monde found, seem largely unconcerned. Meta deleted several pages tied to the Sri Lankan networks, "but since then, little or nothing has changed." When contacted, Meta merely referred the paper to its moderation rules — rules that, Le Monde wrote, "appear to be loosely enforced," noting that Meta's own tools showed Sri Lankan pages running political advertisements banned under the network's European rules without being moderated.
The industry's own conscience isn't exactly reassuring either. "We do not work with pages that clearly cross certain red lines, particularly when it comes to hateful, discriminatory, or overtly extremist political propaganda content," Timothé Bernines, founder of Growlever, a French audience-selling company, told Le Monde — a statement that reads less like a principle and more like an admission of where the bar sits.
The officials tasked with defending democracies from this stuff sound genuinely alarmed. "Profit-driven interference, we see it a bit like a tsunami that is on its way," Marc-Antoine Brillant, director of Viginum, the French government's agency for protection against foreign digital interference, said at a mid-June press meeting. "Many platforms have set up a system to pay their creators. It's massive and it creates a threat profile that is somewhat different [from state-sponsored disinformation actors]."
Closer to this story's home turf, AAP FactCheck has separately debunked "365 Daily News," another Facebook page targeting Australians — this one managed by users in Vietnam — that fabricates quotes from Australian politicians and funnels readers to an ad-laden website, with near-identical posts duplicated across a network of more than a dozen foreign-run pages. The Aussie News Today case isn't an outlier; it's a template.
We've covered this ecosystem's other tentacles before: OpenAI's takedown of a fake Russian think tank run under a false identity, Australia's plan to force AI giants to prove their safety systems actually work, and the awkward episode when Sam Altman sat down with Australia's Deputy PM knowing nothing about a Medicare data breach. The through-line is the same: AI has industrialised deception faster than anyone has industrialised defence.
How to not get fooled
Before writing this, we opened the AAP factcheck and re-verified each of the four incidents, the SynthID verdicts, and the Sri Lanka page-transparency listing ourselves — and we live-checked every link in this article. You can run most of the same checks in about ninety seconds, and you should:
Read the page, not just the post. On any Facebook page, open "Page transparency" and look at where its managers are based. A page screaming about Newcastle that is managed from Colombo has told you everything you need to know.
Stare at the small text. AI image generators still botch writing — uniforms, signs, number plates, logos. The Newcastle crash photo fell apart on the lettering. Zoom in before you share.
Cross-check the story, not the image. Real events will be covered by real outlets with real bylines. If only one page on earth has the "photo" of a major incident — and ABC News, 9News, and everyone else somehow missed it — that is not a scoop. That is a fabrication.
Distrust the format. "BREAKING" in all caps, no named reporter, no original footage, posted by a page you've never heard of: that is the native camouflage of this industry.
The verdict
Let's give credit where it's due, then take it right back. Google's SynthID Detector is a genuine public good — free, fast, and in this case it did exactly what it promised, six days after launch. And Meta's page-transparency panel worked too: the Sri Lanka listing was sitting there in public the whole time, waiting for anyone to look.
But transparency without enforcement is theatre, and detection without consequences is a parlour trick. Meta knew — its own transparency tools showed where these pages are run, its own ad tools showed them buying political ads its rules forbid — and the enforcement was, by Le Monde's careful phrasing, shoddy. A page faking breaking news about a dead soldier stayed up long enough for a national wire service to have to debunk it. That is not a moderation system. That is a suggestion box.
And the deeper problem is the business model humming underneath all of it. Platforms pay for engagement, engagement rewards outrage, and outrage is cheapest to manufacture with AI. Until that incentive changes, fact-checkers will be bailing out a boat that Meta and Google keep drilling holes in — one with a detector, the other with a dashboard, both grading their own homework.
The next fake breaking-news photo is already being generated. The only question is who catches it first — and whether anyone with the power to act actually does.
Sources: AAP FactCheck (debunk published 13 October 2026); Le Monde, "The lucrative business of reactionary AI-generated posts on Facebook" (19 September 2026), reporting on Reset Tech's September 9 study; Google DeepMind's SynthID Detector launch announcement; ABC News and 9News, as cited by AAP.
Daniel Rao is a tech journalist with TechPulse USA covering AI, social platforms, and digital safety — the beats where Silicon Valley's promises meet everyone else's reality.
Aussie News Today fake-news FAQ
What is the "Aussie News Today" page?
A Facebook page billing itself as an Australian breaking-news source, posting multiple times a day about real Australian events — run by a user in Sri Lanka per Meta's page-transparency panel.
How was the fakery detected?
AAP FactCheck checked four posts against real footage and maps, then ran the images through Google's SynthID Detector, which confirmed all were created or edited using Google AI tools. AI image generators still botch writing — uniforms, signs, number plates, logos.
What is SynthID?
An invisible watermark embedded in AI-generated media at creation; Google's detector reads it back. It only catches watermarked content — a "not detected" result proves nothing.
Who profits from this?
Le Monde/Reset Tech traced a profit-driven industry — one Sri Lankan network alone ran 1,340 pages with 36M reactions; engagement earns ad revenue or traffic to ad-stuffed sites. Meta deleted several pages tied to the Sri Lankan networks, "but since then, little or nothing has changed."
Comments
Post a Comment