OpenAI and Meta Told Federal Judges People Can't Spot AI Images. Here's How to Check a Photo Anyway.

What OpenAI and Meta told the judges

The Advisory Committee on Evidence Rules writes the rules federal courts use to decide what evidence is trustworthy. Chaired by U.S. District Judge Jesse Furman of Manhattan, it has been working since 2023 on how courts should authenticate evidence when one side says it's a deepfake (Investing.com).

On October 15, it heard from two of the companies building the technology:

• OpenAI: Shira Anderson, OpenAI's policy counsel, told the panel that AI images continue to improve in quality and that she doesn't believe a scientific method to detect them will exist in the future.
• Meta: Devin Norder, a director of product management at Meta, said media from recent AI models cannot be distinguished by humans from authentic content.

Judge Furman noted that, despite the academic warnings, the issue hasn't come up often in real cases yet. A Federal Judicial Center survey in March found that only 15 of 931 responding judges had seen a party challenge evidence as a deepfake. New laws in California and the European Union require watermarking to help detect deepfakes, but a watermark only helps if it survives: screenshots, crops and re-uploads can strip it.

Why this matters outside a courtroom

Judges can call expert witnesses. You usually have a few seconds and a phone screen. The same images the companies are describing show up where they cost people money and trust:

• A dating match whose photos are of a person who doesn't exist (dating apps)
• A "celebrity" or "investor" account selling a crypto opportunity (investment scams)
• A dramatic news photo shared thousands of times before anyone checks (is this news photo real?)
• A marketplace listing with a photo of an item nobody photographed (fake marketplace photos)

If the people who build these models say a careful look isn't enough, a quick look on a dating app certainly isn't.

Does this mean AI detectors are useless?

No, but it's a fair warning against treating any single test as proof, and that includes ours. OpenAI's point is that no method will be bulletproof. A detector still adds something your eyes can't: it looks for statistical patterns in the pixels rather than for six-fingered hands. That's why Faux Spy gives a confidence score instead of a flat yes or no, and why we say every result is one signal, not a verdict.

The strongest check is several weak ones together: a detector score, the image's history, and whether the story around it holds up.

How to check a photo when your eyes aren't enough

1. Ask where it came from. Who posted it first, and when? A news photo should trace back to a news organization; a dating photo to an account with some history.
2. Look for Content Credentials. Some AI tools and cameras attach a record of how an image was made (what Content Credentials are). A screenshot or re-upload usually strips it, so a missing record proves nothing.
3. Run a reverse image search. It finds real photos stolen from someone else, which an AI detector won't flag (when to use each).
4. Run it through a detector. Upload it at the top of this page or at fauxspy.com/check, or right-click it with the Faux Spy extension on a computer. Read the confidence score, not just the label (how detection works).
5. If money is involved, verify another way. Call a number you already have, or meet on a live video call before sending anything. Urgency is the scammer's favorite tool.