Apple Rewrites App Review Rules to Hit AI Nudify and Bait-and-Switch Apps

Apple has tightened the App Store rulebook on AI-made sexual deepfakes. In an October 16, 2026 update to the App Review Guidelines, Apple made two changes: Guideline 1.2 now explicitly treats nonconsensual intimate imagery, including imagery "generated using AI tools," as content apps must proactively prevent and remove, and a new Guideline 2.5.19 prohibits hidden, dormant or undocumented features. The second rule is aimed at a specific trick, where an app passes review looking harmless and later switches on a nudification feature. Apple published the changes in its developer news.

This post sticks to what Apple has said. We have not tested any app against the new rules, and Apple has not named specific apps in the announcement as we read it.

TL;DR: what changed and who is affected

What exactly does Guideline 1.2 now say?

Apple's release note for 1.2 reads: "Clarified app content and app types that are subject to this guideline, and potential Developer Code of Conduct violation for failing to implement appropriate safeguards or remediating violations after notice."

The accompanying news post goes further. Apple says content moderation requirements now explicitly cover nonconsensual intimate imagery, "including imagery generated using AI tools," as content apps must proactively prevent and remove from their platforms. Developers who fail to implement adequate safeguards, or do not remedy violations after notice, risk violating the Developer Code of Conduct and being removed from the App Store.

Two words matter here: proactively and safeguards. Guideline 1.2 already governed apps with user-generated content, requiring ways to report and block abusive material. The update says AI-generated sexual imagery of real people sits inside that duty, and that reacting only after a complaint is not enough. Apple also restates that pornographic material has long been prohibited, which it says includes apps that generate nudity or sexually explicit imagery, such as "nudification" apps that create sexually explicit images of real individuals without consent.

What is the new Guideline 2.5.19?

The release note says 2.5.19 is a "new guideline [that] clarifies long-standing policy on hidden, dormant, or undocumented features, formerly covered in Guideline 2.3.1."

In the news post, Apple describes the target: apps that misrepresent what they do during App Review and then change after approval, for example by adding features or functionality that were not in the version submitted for review. It names a specific pattern: apps that "misrepresent their purpose during review and only reveal harmful capabilities — such as 'nudification' — after approval," a tactic commonly called bait-and-switch. Apps found doing this are subject to removal from the App Store and from the Apple Developer Program.

This matters because AI apps are unusually easy to change after approval. A photo app can ship a harmless editor and later enable a server-side model that does something else, with no new binary. Server-side switches, remote config flags and prompt changes all alter behavior without going through review. Rule 2.5.19 puts the burden on the developer to describe real capabilities, including dormant ones.

Why now? The timeline behind the update

The change follows a year of pressure on Apple over nudify apps:

• January 27, 2026: The Tech Transparency Project released a report concluding that nudify apps were widely available and easy to find on the App Store, as 9to5Mac reported.
• April 15, 2026: A new report claimed App Store search suggestions and ads steered users to nudify apps, per AppleInsider.
• July 17, 2026: The San Francisco City Attorney's office sent cease-and-desist letters demanding that Apple and Google pull AI nudify apps. 9to5Mac reported that eight of the 13 apps investigated were on the App Store. Apple said it had removed three of them and was in the process of terminating their developer accounts, and was in contact with four others about policy violations.

Apple also said in an earlier statement, after the Tech Transparency Project findings, that it had removed 15 identified apps and contacted developers of six others, giving them 14 days to fix issues or risk removal.

Read against that timeline, the October update is Apple writing its enforcement into the rules, so that it can point to explicit text rather than general content policy when it removes an app.

How does this connect to the wider deepfake fight?

The App Store is one chokepoint among several. Platforms and courts are all dealing with synthetic media at once. We have covered celebrity AI deepfakes on Instagram and YouTube likeness detection, the question of how courts should handle fake evidence in OpenAI and Meta's comments to the judiciary on rule 901(c), and the problem of synthetic political content in AI-generated campaign ads.

Distribution control is different from detection. Detection tries to tell whether an image is fake after it exists. An app-store rule tries to stop the tool from shipping to millions of phones at all. It is blunt, but it works at the scale of one company's review process. The weakness is the same as always: review checks a snapshot, and server-side AI behavior changes later. That is exactly the gap 2.5.19 tries to close with a rule rather than a technical control.

Apple's own AI ambitions are also relevant. The company is shipping more on-device intelligence and tying it into the OS, as in our coverage of Apple's full-disk-access question for AI agents and the Apple Intelligence China rollout claims. A platform that wants to be trusted with AI features has an incentive to look strict about abusive ones.

What should developers do?

If you ship anything that generates or edits images of people, treat these as a checklist:

1. Block at generation time. Add classifiers or policy checks that reject sexual-content requests involving real faces, and test them. "Safeguards" is the word Apple used.
2. Describe every capability in your App Review notes. Dormant features, feature flags and server-side modes count under 2.5.19. If it can be turned on later, say so.
3. Version your server-side behavior. Keep a log of what models and prompts were live when, so you can show App Review or a regulator what changed.
4. Have a fast takedown path. The text penalizes failing to remedy violations after notice. Build a reporting channel and a response clock.
5. Plan for audit. If an app is flagged, expect Apple to ask what the app can do, not what the review build did.

For users and parents: the practical effect is that apps promoted with innocuous descriptions are more likely to be removed once a hidden capability surfaces, but the rules are enforced after the fact. Do not treat App Store presence as proof that an app is safe. If someone has made sexual imagery of you without consent, Apple's report tools and the platform where the image is shared are the first stops, and many jurisdictions now have specific laws.

How will this play out for AI app makers?

Expect three effects. First, review friction: apps with image generation, face swap or "AI avatar" features will likely face more questions about filters and about what the backend model can do. Second, a compliance paper trail will become part of shipping, much as privacy labels did. Third, enforcement will probably arrive through high-profile cases first, because Apple's earlier removals followed press reports and official letters rather than systematic sweeps.

For small teams there is a quieter risk: a legitimate editing app built on an open image model can drift into territory the rule covers if users find prompts that bypass its filters. The guideline's language about remediating violations after notice suggests that a fast, documented fix is the difference between a warning and removal. Teams that rely on third-party model APIs should check the provider's own safety settings, since Apple will hold the app developer accountable either way.

What is still unclear?

• Enforcement numbers. Apple's post, as summarized from its developer news page, does not give removal counts or a date by which existing apps must comply.
• Scope of "proactively." The guideline does not say which filters count as adequate. Expect case-by-case decisions.
• Server-side changes. 2.5.19 speaks of features added after approval. How Apple will detect server-driven behavior change is not described.
• Other stores. Google was also named in the San Francisco letters, and we did not find a matching Play policy change in this research.

The primary text lives in Apple's App Review Guidelines, which you should read in full before relying on any summary, including ours.

Related reading

• Celebrity AI deepfakes on Instagram and YouTube likeness detection
• OpenAI and Meta on deepfake evidence and rule 901(c)
• AI-generated campaign ads: 220 ads, 137 million, undisclosed
• Apple full disk access and AI agents
• Apple Intelligence in China: claimed vs verified
• Meta AI in Brazil: Lula posters blocked, Flávio posters generated
• Official: Apple developer news, App Review Guidelines

Guideline text and enforcement details are accurate as of October 17, 2026 and may change.