A warning for anyone who donates online, follows influencers, or builds trust through a screen: the tools to fake an entire charity now cost almost nothing and leave almost no trace. That’s the takeaway from an investigation making the rounds on Hacker News this week, where an ABC NEWS Verify report on Australian influencer Lily Jay exposed what reporters call a “tangled web of AI manipulation” behind a foundation that solicited real donations from millions of followers.
Here’s what happened.
What the investigation found
Lily Jay has nearly 3 million Instagram followers, built around her public conversion to Islam and the work of the Lily Jay Foundation. According to Hacker News, ABC NEWS Verify dug into the foundation’s charitable claims and found them riddled with fabrication.
The specifics are worth spelling out:
- The woman announcing a new orphanage in one viral video isn’t the real Lily Jay. She’s an AI-generated fake. So are the children holding lollipops around her, and the foundation banner behind her.
- A telltale error gave it away: an extra “L” appeared on the back of a worker’s shirt. Garbled text is a classic fingerprint of generated images.
- A May press release claimed Lily Jay won a “2026 Austral-Global Excellence Award for Humanitarian Leadership.” Analysis confirmed the award photos carried ChatGPT’s SynthID watermark, meaning the images were machine-made. Investigators couldn’t find any trace of the award existing outside the foundation’s own materials.
- The PR firm behind the release, Real Media Group, listed Lily Jay as both a “client” and a “co-founder” on its own site.
In Uganda, where the foundation claims to run an orphanage, operating one without government registration is illegal. The registration bureau initially confirmed no orphanage existed under the foundation’s name. A registration appeared only days after ABC sent questions, and it’s listed as “not compliant.”
Why this matters
What stands out here isn’t a single deepfake. It’s the blend. The most dangerous part of this story is how AI-generated footage was spliced together with real clips, making the whole thing hard to untangle. A sign on a truck flickers over an arm, betraying the edit. A face is too smooth. But most viewers scrolling past won’t catch any of it.
This is a shift in the threat model for online trust. Faking a charity used to require staged photos, hired actors, or at least stolen images from somewhere real. Now a single person can generate the founder, the beneficiaries, the awards, and the press coverage from a prompt box. The cost of manufacturing credibility has collapsed.
Tim Costello, former CEO of World Vision Australia, put the older version of this problem plainly in the report: “Pictures of kids in orphanages tug at a donor’s heart like nothing else.” Generative AI just industrialized that manipulation.
What to watch for
The defensive playbook is changing, and donors carry more of the burden now. A few practical checks before you send money or trust an aid campaign:
- Verify registration independently. Legitimate charities are registered with a government body. Look it up yourself rather than trusting a logo.
- Corroborate on the ground. Real aid work leaves a footprint: partners, officials, and other operators who’ve heard of it. In this case, humanitarian sources in Gaza had never heard of the foundation’s claimed bakery, and investigators couldn’t geolocate it.
- Watch the small details. Misspelled text, flickering logos, and unnaturally smooth faces still leak through. They won’t forever, so don’t rely on them alone.
- Distrust self-referential proof. An award only the recipient can confirm isn’t an award. A PR firm that also employs its “client” isn’t independent coverage.
The uncomfortable part is that detection tools like SynthID watermarks caught the fake images this time, but only because investigators knew to look. Watermarking isn’t universal, and plenty of models don’t apply it. Red Crescent told ABC it has seen a rise in these kinds of online campaigns, which suggests this case is an early signal, not an outlier.
Expect more of this, aimed at causes that pull hardest on your empathy. The full investigation, with side-by-side breakdowns of the manipulated footage, is worth reading at the original source.