Aug. 16, 2026・Issue #126

Let's talk about what's going on with Snopes: the newsroom, the products, the people and everything and anything that makes Snopes, Snopes. Here's your exclusive behind-the-scenes look into what we do, available only for paying members of our fact-finding community:

Over the years, Snopes has covered plenty of misinformation and disinformation about Dr. Anthony Fauci, who served as chief medical adviser to U.S. presidents Donald Trump and Joe Biden during the COVID-19 pandemic.

Most of the claims about him were politically motivated, particularly by conservative media outlets and politicians.

Snopes has covered many of the misleading rumors, incorrect quote attributions and claims that are outright false. Below are our roundups of Fauci rumors:

We need your continued support to continue our fact-finding mission.

View more options

Writing this story — simply a laundry list of techniques we use to spot AI images — wasn't as easy as you might think. When generative AI image models first started appearing, there were a number of dead giveaways. But the signs have become harder to spot as models get more advanced.

For instance, text used to be a go-to way to tell whether an image was real — if it's gibberish, for example, that's a clear sign of an AI image. These days, it's still worth checking, but models have become significantly more precise.

That means we've had to come up with different methodologies to suss out AI-generated media. That made the writing process even longer. As new, more-photorealistic models went public, I found myself adding new tips and tricks and caveats to previous ones.

To give an example: When I started writing this story, SynthID was a relatively new tool from Google that we were just beginning to use in our fact checks. While we do not recommend using SynthID as a sole source of information, you'll find it in practically all of our stories debunking fake images because it's convenient.

I hope that you find all of the tips in this article helpful. But I picked two to highlight. First: Be practical and think critically while browsing online. If you get in the habit of doing this, I know from personal experience that you'll catch lies and fabricated media more often.

Second, when it comes to AI-generated media specifically, home in on complex shapes, like hands hair and ears, which AI models still struggle with. If they're misshapen or blend into one another, you're probably looking at an AI image.

Overheard: "i hear a yoga sesh a day keeps the 30-50 feral hogs away"

— Reporter Emery Winter

Earlier this month, we investigated the claim that AI companies were scanning and destroying books to train their models, including rare books.

We pored over hundreds of pages of Anthropic's internal documents unsealed amid a copyright lawsuit to confirm the company was slicing the spines off books it bought in bulk and "destructively scanning" them to train its AI model, Claude.

The documents proved Tom Turvey, the project lead, agreed to prioritize "less common books," without defining the phrase. Still, it appeared to support the claim that Anthropic was targeting possibly rare books.

After publication, a reader emailed us to say our report contradicted Anthropic's assertion to The Guardian that it did not scan "rare or antiquarian books," though the company didn’t say how it defined those terms. We pressed a spokesperson at Anthropic to provide clarity. At first, the spokesperson sent us the same statement given to The Guardian. In the exchange that followed, we insisted that the company define the term.

The spokesperson eventually confirmed that the company made a distinction between "rare and valuable" — that is, akin to collectibles — and harder-to-find books, like professional or reference books. In practice, this could give the company a wide berth to target rare books that do not fall into the category of collectible objects. This clarification allowed us to maintain our "mostly true" rating.

Last month, a reader flagged passages by a former Snopes writer that appear to contain writing generated by artificial intelligence software without disclosure. That message prompted our own look into the former writer's work, and we identified 13 articles published in 2024 that appear to violate our editorial policies. A letter from me on our site has more information.

Now is a watershed moment for the publishing industry when it comes to AI. We appreciate your continued support in enforcing the highest of high ethical standards: We require original writing for publication and our fact-checkers to always disclose their methods of researching and writing.

Our credibility is our lifeline. Snopes does not take for granted the trust of readers like you: people who believe in the power of quality, human-made journalism.

— Jessica Lee, executive editor of Snopes

Oh Lord, AI Reporters Are Actually Breaking Big News
— Kate Knibbs, Wired

Revising the Ethics Code
— Liz Skalka, Columbia Journalism Review

Ida B. Wells' Family Is Debunking a Viral Claim About Nolan Wells
— Brandon Tensley, Capital B

Thanks for reading this edition of the Snopes Digest. We send new issues of the newsletter every month, so please add this email address to your white list and keep an eye out for the next issue.

Have feedback or a story idea you want to share?