Deepfake Watermark Law Passed in 28 Jurisdictions: What Provenance Labels Can and Cannot Do

A short video appears in a feed: a public figure appears to say something shocking. In the past, viewers had to guess whether it was real. In the scenario that follows, a new rule changes the starting point. Across 28 fictional jurisdictions, laws now require that synthetic or heavily edited media carry a cryptographic provenance mark, a label built into the file that says where it came from and how it was altered.

Launch edition note: the laws, the jurisdictions and the numbers here are fictional. This is an illustrative scenario about how provenance labels work; it names no real law or company.

Two ideas that are often confused

People use "watermark" for several different techniques, and the differences matter.

• Provenance metadata is a signed record attached to a file. It states the device or software that created it and a list of edits. A digital signature makes tampering with the record detectable.
• Invisible watermarks are patterns hidden inside pixels or sound that software can detect. They can survive some copying but not all.
• Visible labels are notices displayed to the viewer, such as "AI-generated."

The scenario's rules require provenance records and encourage invisible watermarks as a backup. Visible labels are required where the content could mislead.

How a viewer would check

When a device or app supports it, viewing a file shows a small information icon. Opening it displays the signed record: which tool made the image, whether it was edited, and which organization vouches for the signature. If the record is missing or the signature does not match, the viewer sees a warning rather than a verdict. The idea is to give people more information, not to tell them what to believe.

"A provenance label is a receipt. It tells you where something was bought, not whether it was worth buying." — a policy analyst at a digital rights group (fictional)

What labels can do

1. Raise the cost of deception. A forger now has to strip or fake a signature, which takes effort.
2. Help honest creators. Photographers and news organizations can prove that an image came from a camera and was not altered.
3. Support investigators. Fact-checkers can start from a file's history rather than a hunch.

What labels cannot do

This is where careful reading matters, and where the scenario's own regulators are frank.

• Stripping. Metadata can be removed by taking a screenshot or re-encoding a file. A missing label proves nothing.
• Unsupported tools. Software outside the 28 jurisdictions, or open tools that simply ignore the rules, will produce unmarked content.
• Truth versus origin. A real photograph can be captioned falsely. A synthetic image can illustrate a true event. The label speaks only to how the file was made.
• False confidence. If viewers come to trust any file with a mark, a forger who obtains a valid signature gains credibility.

Enforcement and fairness

Rules are only as good as their enforcement. The scenario describes penalties for platforms that knowingly distribute unlabeled synthetic media, but notes that detection at scale is imperfect. Civil-liberties groups raise a different concern: that provenance requirements could expose the sources of whistleblowers or burden small creators. The rules include exemptions for anonymity-protecting tools, and regulators say they will review the balance.

What viewers can do

• Check for a provenance icon where available, but do not treat its absence as proof of fakery.
• Look for the original source of a clip before sharing.
• Be skeptical of content designed to provoke outrage.
• Look for corroboration from independent outlets.

Lessons from earlier labeling efforts

Labels have a mixed history in other fields. Nutrition panels changed what companies made because people could compare products at a glance, but warning text that grew too long was often ignored. The scenario's designers say they tested the provenance icon with ordinary users and found that a simple, consistent symbol worked better than detailed explanations. They also found that people overestimated what the icon meant, which is why the information panel states plainly that a mark describes how a file was made and does not vouch for what it says. Education campaigns are planned alongside the rules, because a tool that readers misunderstand can do as much harm as having none.

What to watch next

Key indicators include how many devices and apps actually display provenance information, whether platforms preserve the marks when files are shared, and how quickly misuse is penalized. The scenario's regulators plan a review after one year, and say they will publish findings on how often labels were stripped.

The rule is a useful tool, but not a final answer. The best defense against deceptive media will probably remain a mix of technology, institutions and readers who ask where a clip came from before they pass it on.