You use AI tools somewhere in your production work, and you post to Instagram and Facebook. Will Meta label the post for you, or is disclosure your job?
The question, and what this piece can and cannot answer
This is a reading of Meta’s published statements. Nobody uploaded anything to see what appears in the app. There are no screenshots of labels and no observed behaviour. Where a sentence says what the label does, it is reporting what Meta says it does.
The thesis, as a reading of Meta’s text: Meta says the label is driven by signals it detects plus self-disclosure by the person posting. That leaves your own disclosure practice as the part you control. This piece makes no claim about how much AI-assisted content goes unlabeled, because no source read supports a number.
On dates: the newest primary statement found is from February 19, 2026. The detailed statements are from 2024.
The signals Meta says it reads, in its own words
Meta names four inputs across its statements. The table keeps each attached to its source and date.
| Signal | What Meta says about it | Source and date |
|---|---|---|
| “AI generated” information in the C2PA and IPTC technical standards | Meta says it is building tools that can identify these invisible markers at scale, so it can label images from other companies’ tools. | Meta newsroom post on labeling AI-generated images, February 6, 2024 |
| Invisible watermarks and embedded metadata on Meta AI images | For photorealistic images made with its own Meta AI feature, Meta says it applies visible markers, invisible watermarks and metadata, and labels them “Imagined with AI”. | Same post, February 6, 2024 |
| Self-disclosure by the person posting | For video and audio, Meta says it is adding a feature for people to disclose so it can add a label. Later statements name self-disclosure as a trigger too. | February 6, 2024; April 5, 2024 post; February 19, 2026 statement |
| “Industry-shared signals” | Meta says labels are based on detecting industry-shared signals of AI content or people self-disclosing. The September 2024 update says the label shows which of the two it was. | April 5, 2024 post; update of September 12, 2024 |
The February 2024 post also names companies. Meta says it is building tools so it can “label images from Google, OpenAI, Microsoft, Adobe, Midjourney, and Shutterstock as they implement their plans for adding metadata to images created by their tools.” Read that as Meta’s stated intent in 2024, not a confirmed current list of tools that embed the signals.
The newest statement, February 19, 2026, sits in a post about the 2026 US midterms, in a section on identifying AI-generated content beyond political ads. It says Meta uses “a combination of industry standards and technology (such as C2PA)” to help identify content generated or edited using AI outside of ads, and that it displays an “AI info” label for content it detects was generated by an AI tool. It also says it shares whether the content is labeled because of industry-shared signals or because someone self-disclosed.
Video and audio: the case where Meta says disclosure is required
This is the one place where Meta’s text moves from “we detect” to “you must tell us.” The reason Meta gave in February 2024 was practical. Companies were starting to include signals in their image generators, but had not started including them in AI tools that generate audio and video at the same scale. So Meta said it could not yet detect those signals and label that content from other companies. That was its 2024 position, and nothing read here says whether it is still true.
What Meta says is required, in both the 2024 post and the 2026 statement, is close to identical. People must use the disclosure and label tool when they post organic content with a photorealistic video or realistic-sounding audio that was digitally created or altered, and Meta says it may apply penalties if they fail to. Because the 2026 statement repeats the requirement in nearly the same words, this is the best-dated claim in the piece.
Two limits matter. First, the pages read do not state a requirement to use the tool for images, and this piece leaves it there rather than extending the rule. Second, “may apply penalties” is Meta’s wording. The pages read do not say what a penalty is, and this piece does not guess at one.
What changed in 2024, in order
In order:
- April 5, 2024. Meta announced it would start labeling organic AI-generated content in May 2024, and would stop removing content solely under its manipulated video policy in July. Labels would be based on detecting industry-shared signals or on self-disclosure.
- May 10, 2024 update. Meta clarified that the plan applies to organic content.
- July 1, 2024 update. “Made with AI” became “AI info.” Meta’s stated reason: some content with minor AI modifications, such as retouching tools, carried industry standard indicators and was labeled “Made with AI,” and those labels did not always match people’s expectations. Meta says people can click “AI info” for more information.
- September 12, 2024 update. For content Meta detects was only modified or edited by AI tools, the “AI info” label moves to the post’s menu. Content Meta detects was generated by an AI tool still shows the label, along with whether it is there because of industry-shared signals or self-disclosure.
The practical consequence of the last entry is a distinction Meta itself draws: content it detects was “generated” versus content it detects was “only modified or edited.” The update, as worded, says the changes “will begin next week,” so it describes a rollout on that date, not the app today. The pages read do not say how Meta decides where a given asset falls between the two, and this piece does not guess.
One more dating detail. The April page’s HTML metadata records a modification date of October 23, 2025, later than the last visible dated update from September 12, 2024. The page does not say what was edited. That was re-checked in this session.
What Meta says it cannot catch
Meta is direct about the limits in the February 2024 post. It says: “it’s not yet possible to identify all AI-generated content, and there are ways that people can strip out invisible markers.”
The same post says Meta is working on classifiers that could automatically detect AI-generated content even when it lacks invisible markers. It also describes watermark research, a technology it calls Stable Signature, which integrates the watermark into the image generation process for some types of image generators. Meta describes classifiers as under development in 2024, and the pages read do not report whether they are now in use.
The consequence for you is that the absence of a label tells you nothing about whether content was made with AI. Meta says it cannot identify all of it.
Working through Meta’s stated rules (reasoning, not observation)
What follows is reasoning from Meta’s wording. These are readings of the text, not observed results, and there is no example asset because nothing was tested.
- If an asset carries C2PA or IPTC “AI generated” information, then Meta says it has been building tools to detect that at scale. (Meta, February 6, 2024.)
- If an export or editing step removes that metadata, then Meta’s own text says people can strip out invisible markers and that it cannot identify all AI-generated content, so a missing label proves nothing. (Meta, February 6, 2024.)
- If the post is a photorealistic video or realistic-sounding audio that was digitally created or altered, then Meta says you are required to use the disclosure and label tool for organic content, whether or not any signal is present in the file. (Meta, February 6, 2024, repeated February 19, 2026.)
- If the asset was only edited with AI tools and Meta detects that, then Meta says the “AI info” label sits in the post’s menu rather than on the post. (Meta, September 12, 2024.)
A disclosure checklist, as questions to answer before you post. This is not legal advice.
- What type of media is it: image, video, audio, or text?
- Was it generated by an AI tool, or only edited by one?
- Does it depict something photorealistic, or contain realistic-sounding audio?
- Is it organic, or is it an ad?
The last question matters because ad rules are separate. Meta’s February 2026 statement addresses political ads on their own, and this piece does not go further into ad rules. A paid-partnership label is a separate matter too, covered in the paid partnership label and the FTC disclosure standard.
What could not be confirmed as current
The February and April 2024 posts and the September 12, 2024 update are more than two years old as of September 2026. The February 19, 2026 statement is the newest primary statement found, and the only one inside two years. These are the most recent statements found, not a guarantee of how the label behaves today.
Meta’s Transparency Center page on labeling AI content and the Instagram help center were not readable in this session. The Transparency Center page returned HTTP 200, but its text is script-rendered and only a page title came through. Nothing on either page is cited here.
This piece also does not cover how the label affects reach, ranking or distribution, because the pages read do not say. For what Instagram has and has not published on that subject, see what Instagram has published about ranking, and what it has not.
FAQ
Does Instagram label AI content automatically?
Meta says it adds the “AI info” label when it detects industry-shared signals, such as C2PA and IPTC “AI generated” information, or when the person posting discloses (Meta, February 6, 2024; April 5, 2024 with its updates; February 19, 2026). Meta also says it is not yet possible to identify all AI-generated content, so no label is not proof either way.
Do I have to disclose AI use myself?
Meta says people must use its disclosure and label tool when posting organic content with a photorealistic video or realistic-sounding audio that was digitally created or altered, and that it may apply penalties if they fail to (February 6, 2024, repeated February 19, 2026). The pages read do not state the same requirement for images.
Sources
- Meta, “Labeling AI-Generated Images on Facebook, Instagram and Threads,” published February 6, 2024. Fetched September 24, 2026, HTTP 200.
- Meta, “Our Approach to Labeling AI-Generated Content and Manipulated Media,” published April 5, 2024, with updates through September 12, 2024. Fetched September 24, 2026, HTTP 200.
- Meta, “How Meta Is Preparing for the 2026 US Midterm Elections,” published February 19, 2026. Fetched September 24, 2026, HTTP 200.
