Quick answer: Meta announced that starting May 2026, AI-generated content on Facebook and Instagram carries mandatory labels. Starting July 2026, deepfake videos will no longer be removed solely because they are manipulated media. Removal continues only when a deepfake separately violates another Community Standard, such as voter interference or non-consensual intimate imagery.
For years Meta's response to a deepfake video that violated its manipulated-media policy was removal. A deepfake of a politician saying something they did not say, a fabricated event, a synthesized clip designed to mislead: the platform's stated response was to take it down.
That changes in July 2026. Meta announced a policy shift in which deepfake videos, by themselves, will no longer be removed. The new default is a label. Removal continues only when a deepfake also violates a separate Community Standard, such as non-consensual intimate imagery, voter interference, or coordinated inauthentic behavior. For everything else, the video stays up with a "Made with AI" or equivalent label attached.
This post walks through what the policy actually says, what it changes for users on Facebook and Instagram, and what it means for the detection approach our Facebook detection guide and Instagram detection guide recommend. For the underlying technical grounding on how deepfake video gets generated and labeled, see the pillar guide on what a deepfake actually is.
What Meta Actually Announced
The policy was announced by Monika Bickert, Meta's Vice President of Content Policy, and lands in two phases.
May 2026: Mandatory AI-content labels begin. Facebook and Instagram will display a visible label on content the platform identifies as AI-generated or AI-manipulated. The label depends on a combination of C2PA Content Credentials embedded by source AI tools, automated detection systems, and creator self-disclosure.
July 2026: Meta will stop removing deepfake videos solely on the basis that the content is manipulated media. The "manipulated media" provision of the Community Standards previously triggered removal; in July it triggers labeling instead. Removal continues for any deepfake that separately violates another standard: non-consensual intimate imagery (covered by the TAKE IT DOWN Act compliance path as well), voter interference, threats of violence, scams, fraud, coordinated inauthentic behavior, or harassment.
Meta characterized the shift as favoring transparency over removal as a means of preserving freedom of expression. The reasoning given: a label informs the viewer that the content is synthetic without removing the speech, while a takedown is a heavier response that the platform argues is disproportionate for manipulated media that does not separately cross another harm line.
What Changes for Users
The practical effect depends on what kind of deepfake you encounter and what you would have expected the platform to do about it.
Non-consensual intimate imagery: no change. NCII still triggers removal under both Meta's NCII policy and the federal TAKE IT DOWN Act 48-hour requirement. The filing flow in our per-platform takedown notice guide continues to work the same way on Meta platforms.
Voter-interference deepfakes: no change. A deepfake of a politician designed to mislead voters around an election still gets removed if it violates the voter-interference provision. This was the highest-profile category of deepfake takedown historically, and it stays under removal jurisdiction.
Scam, fraud, and impersonation deepfakes: no change. A deepfake celebrity endorsing a fake crypto platform, a fake executive demanding a wire transfer, an impersonation account: all still get removed under separate policies. The celebrity crypto deepfake guide and the CEO deepfake fraud post cover the patterns that continue to trigger takedown.
Everything else: label only. The deepfake category that loses removal protection is the residual: synthetic video that is not NCII, not voter interference, not scam, not impersonation, not harassment, just a misleading or fabricated video that depicts something that did not happen. Under the previous policy, that category was removable under "manipulated media." Under the July 2026 policy, it stays up with a label attached.
Two policy phases
May 2026: mandatory AI labels begin on Facebook and Instagram. July 2026: deepfake removal narrows to videos that violate a separate Community Standard. The label becomes the default; removal becomes the exception.
Source: Meta policy announcement, reported May 2026.
Why the Label Will Not Carry the Weight Meta Is Asking It To
The shift assumes the label is a meaningful intervention. The available data suggests it is not.
The label depends on detection that operators bypass. Meta's automatic detection looks for C2PA Content Credentials embedded by source AI tools. Operators strip those credentials with one conversion step before upload, leaving the label dark on most synthetic content the platform encounters. A small ecosystem of "AI label remover" services exists specifically for this workflow. The label fires on cooperating tools used by good-faith creators, and misses the fraud and disinformation cases the policy shift is most consequential for.
The label has poor user engagement when it does fire. Meta's own Q1 2026 Transparency Center data, cited in our Instagram detection guide, showed the AI Info label was displayed on hundreds of millions of Reels with only a small fraction generating any user interaction. A label that the viewer does not see is functionally equivalent to no label, and a label that gets stripped before upload is not displayed at all.
The pattern Stanford Internet Observatory documented for AI-spam Pages on Facebook, with one post hitting 40 million views, does not change under this policy. Those Pages produce content that is misleading or fabricated but does not violate a separate Community Standard. Under the old policy the manipulated-media provision was a potential takedown lever for the worst Page-cluster content; under the new policy it is not, and the label is the only response.
What This Means for You
Three rules that hold for any Facebook or Instagram content you encounter after July 2026.
Treat the label as a positive signal when present and not a clearance when absent. A piece of content carrying an AI label is, by Meta's own classification, synthetic. A piece without a label could be authentic, or it could be synthetic content the detection failed to catch. The label's absence proves nothing.
Use the detection guides for the cases the label misses. The seven signs for Instagram Reels and the seven signs for Facebook video are now load-bearing. The label was always a complement to manual detection; under the July 2026 policy, it becomes a smaller complement to a larger manual detection burden.
Reporting changes too. For deepfakes that fall in the residual category (synthetic, misleading, but not crossing another harm line), reporting no longer produces a takedown by itself. Reports still feed the pattern data Meta's enforcement systems use, and clusters of coordinated synthetic content are reportable under coordinated-inauthentic-behavior, which retains removal authority. For non-NCII reporting flows, the what-to-do-after-finding-a-deepfake guide covers the documentation and reporting sequence.
If a video on Facebook or Instagram caught your attention recently and you want to check whether the account behind it has already been flagged by the Ledger community, paste the URL below.
Think you found an AI video?
Paste the URL and let the Ledger community verify it. Free.
The Bigger Pattern
The Meta policy shift is not happening in isolation. The EU AI Act's Article 50 transparency rules, taking effect August 2, 2026, formalize a labeling regime across all major platforms operating in EU markets. Meta's "labels over removal" posture aligns Meta with that regulatory baseline; the question is whether labels are actually doing the work the regulation assumes they are.
The honest read is that this shift moves more of the detection burden to the viewer. The label is one signal among many, and the platform is now relying on the viewer to weight it correctly. For consumers who have been using the detection guides on this site, the change is incremental. For consumers who were trusting Meta to remove the manipulated content they reported, the change is meaningful. The defense framing has not changed; the platform's share of the defense has shrunk.
Related Posts
- How to Tell If a Facebook Video Is AI-Generated (and Help Your Parents Spot One): the Facebook detection guide that becomes load-bearing under the new policy regime
- How to Tell If an Instagram Reel Is AI-Generated: 7 Signs to Check Before You Share: the Instagram detection guide and the Meta label-failure analysis that the policy shift confirms
- 120 Facebook Pages Were Quietly Running AI Content. One Post Hit 40 Million Views.: the Stanford SIO documentation of the Page-cluster pattern that loses its removal lever under the new policy
- How to File a TAKE IT DOWN Act Takedown Notice on TikTok, Instagram, Facebook, X, and YouTube: the NCII removal pathway that continues to work the same way on Meta platforms regardless of the manipulated-media policy change
