When a student's work is challenged, build a record that holds up.

Academic-integrity panels are being asked to defend AI-related findings on appeal, and schools are taking on new duties to disclose their own AI-generated content. Provenance Radar surfaces the underlying evidence — Content Credentials, metadata, watermark signals, and AI-generation indicators — for both jobs, without reducing either one to a single score.

Scan content

Documents and images · More formats coming soon

Why this matters

Two jobs, one evidence standard

Academic integrity and institutional disclosure look like separate problems, but they run on the same underlying question: what evidence actually supports what you're about to say, in either direction.

Investigating student work

When a submission is disputed, a reviewer needs more than a single AI-detector percentage. Provenance Radar checks each file for Content Credentials, embedded metadata, watermark claims, and AI-generation indicators, and labels every signal by certainty, so a reviewer can see exactly what was found, and what wasn't, before making a judgement call.

Your institution's own disclosure duty

Schools don't just assess AI use in student work, they're increasingly expected to disclose their own. As admissions material, newsletters, and public communications start to include AI-assisted or AI-generated content, a growing number of jurisdictions expect institutions to be transparent about it. The same evidence categories that support a fair review of student work can be turned on your own content before it's published.

What happens if you don't

The cost of skipping the evidence

Defensibility risk

A finding based on one AI-detector score is hard to defend once it's challenged. If a student appeals, or a hearing asks how a conclusion was reached, "the tool said 92 percent" isn't an answer. A decision needs a trail behind it, showing what was checked, what was found, and what was verified versus merely inferred, for a human reviewer to stand behind.

Reputational risk

If your institution publishes AI-generated or AI-assisted content without being able to say so, and that becomes public, it's a harder conversation precisely because your institution also investigates students for the same thing. Getting ahead of your own disclosure question keeps that standard consistent in both directions.

Global regulatory status

This isn't an EU-only rule

AI-content disclosure obligations are landing in one jurisdiction after another. Here's what's already in force, and what isn't yet.

China

In effect

In effect since 1 September 2025

The Measures for Labeling AI-Generated Content require both a visible label and an embedded metadata label on AI-generated text, images, audio, and video.

South Korea

In effect

In effect since 22 January 2026

The AI Basic Act requires a visible label on realistic AI-generated content and an invisible watermark on stylised generated content. Penalties are deferred during a grace period, except for cases of serious harm.

European Union

In effect

In effect since 2 August 2026

Article 50 of the AI Act requires generative AI providers to mark synthetic content as machine-readable, requires deployers to disclose deepfakes, and requires AI-generated public-interest text to be labelled.

United States

Active, state by state

Expanding since August 2026

California's AI Transparency Act took effect 2 August 2026, requiring generative AI providers to offer watermarking and disclosure tools. Most states now have deepfake disclosure laws for political content — there is no single federal standard.

Australia

Pending

No dedicated law yet

Proposed mandatory AI guardrails were shelved in the December 2025 National AI Plan in favour of existing privacy and consumer law. The only date on the books is a narrow automated-decision disclosure duty from 10 December 2026 — not a general AI-content labelling rule.

This is a general summary for orientation, not legal advice — confirm current requirements for your jurisdiction with counsel before relying on it.

How it works

From upload to evidence you can act on

The same four steps, whether you're reviewing a disputed submission or checking your own content before it goes out.

01Upload the disputed file, or a piece of your own published content
02Provenance Radar checks for Content Credentials, embedded metadata, watermark claims, and AI-generation indicators
03Each signal is labeled by certainty — verified, declared, detected, inferred, or unknown — never combined into one score
04Your reviewer, panel, or communications lead weighs the evidence and makes the call

View a sample report

Questions from academic-integrity teams

Evidence, never a verdict

No, it never does. Provenance Radar produces evidence — what's verified, what's declared, what's detected, what's inferred, and what's simply unknown. Weighing that evidence and reaching a finding stays with your reviewer or panel, the same as it would with any other piece of evidence in a case.