Meta says its AI systems detected more than 97% of the child sexual exploitation content. The company removed it or took action in the first half of 2026, before users reported it. It took action against 33.2 million pieces of such content on Facebook and Instagram between January and June. Despite Meta’s progress in detecting harmful content, dangerous AI tools and their ads are still appearing on its platforms.
In an October 7 announcement, Meta introduced new AI systems to detect ads that lead users to child sexual abuse material hosted on external websites. The company is also using AI to flag suspicious ads, test its safety systems, and identify people who return after their accounts have been removed.
This shows that Meta is taking steps to improve its response to online exploitation. However, when people’s safety is at stake, platforms must do more than detect harmful content. They must prevent it from reaching users in the first place.
The issue extends beyond child safety. Meta has also faced scrutiny over ads for AI nudify apps that can create sexual images of real people without their consent. These cases raise concerns about whether Meta’s ad review systems can keep up with the growing misuse of AI tools.
Meta’s 97% Detection Rate Does Not Prove Its Platforms Are Safe
Meta says it detected more than 97% of the child sexual exploitation content it acted on between January and June 2026 before users reported it. In India, the company reported a rate of over 98%, covering 5.3 million pieces of content across Facebook and Instagram.
This shows that automated content moderation helps Meta identify harmful material before users flag it. However, they do not reveal how much illegal content its systems failed to detect, how long it took to remove violating posts, or whether harmful ads reached users before being taken down.
This covers child sexual exploitation content across Meta’s platforms, not just ads. This means they cannot be used to claim that Meta’s ad review system detects 97% of harmful advertisements before they reach users.
Meta has shared figures on the content it removed and the material it detected without user reports. However, these numbers alone do not show how much harmful content escaped detection or how well its moderation systems work overall.
This is an important gap in Meta’s safety claims. Effective content moderation should do more than remove reported violations or detect harmful material after it appears. It should also prevent users from seeing such content in the first place. That includes stopping ads for AI nudify apps and other services that can be used to create sexual images of people without their consent.
Meta Uses AI to Find Suspicious Ads and Expose Hidden Abuse
On October 7, Meta announced new AI tools to detect child sexual exploitation content and identify suspicious advertisements. The company said some offenders were using ads that looked harmless but directed users to external websites hosting illegal material.
Meta calls this tactic “signposting.” Instead of showing prohibited content in an ad, offenders use the ad to direct people to another website where the material is available.
To address this problem, Meta is using a large language model to identify suspicious ads based on their content and purpose. It is also checking the websites linked to ads, rather than relying only on the images and text displayed in them.
Other updates include extra AI checks to catch content missed by existing systems, an AI agent to test its safety measures for weaknesses, and better ways to identify people who return with new accounts after being banned.
These changes target a gap in ad moderation. An ad may look harmless on its own, but the website it links to can reveal its real purpose.
However, the need for these new tools raises questions about Meta’s earlier safeguards. If offenders could use harmless-looking ads to direct users to illegal content, why did the existing systems fail to catch this pattern sooner?
Meta says it expanded its investigation after discovering the tactic. While these updates may help detect such ads, platforms should keep improving their moderation systems before new abuse methods spread, rather than waiting for a problem to emerge.
ALSO READ: Meta Told to Toughen Deepfake Rules as AI Nude Abuse Spreads
Meta Still Struggles to Block Ads for AI Nudify Tools
Meta’s content moderation challenges are not limited to child safety. The company has also faced criticism for allowing ads that promote AI nudify tools, which can create fake sexual images of real people without their consent.
An August 18, 2026, investigation by WIRED found that Meta ran ads for Kromix, an AI image and video app. Some of its promotional material appeared to show sexual deepfakes resembling female US politicians.
One ad showed a woman resembling a prominent US politician before switching to a pornographic video featuring a similar-looking face. Meta removed the ads after WIRED contacted the company about them.
The investigation also found that some ads used innocent-looking images, such as landscapes, to hide sexually explicit content elsewhere in the video. This shows how advertisers can use misleading visuals to get around content review systems.
Meta said it removed more than 340,000 ads promoting nudify apps between November 2025 and January 2026. The company also said it was developing new technology to identify and block ads that try to bypass its rules.
However, the Kromix case shows that ads promoting AI nudify tools could still get through Meta’s review process.
AI moderation remains important because automated systems can scan large amounts of content quickly. But high detection rates alone do not prove that a platform is protecting users effectively.
Moderation systems may catch known violations but miss new tactics, misleading ads, or content that requires a closer look at its context. Meta’s experience with nudify ads highlights why its safeguards need regular testing and independent scrutiny.
The same concern applies to child safety. Platforms must do more than detect known abusive content. They also need to identify new methods used to promote harmful material and stop it from reaching users.
ALSO READ: AI Has Made Fake Nude Abuse Easier to Scale, New Berkeley Report Warns
Meta Should Be Held Accountable for the Ads It Approves
Meta earns money by allowing advertisers to promote products and services across its platforms. Its ad review systems decide which promotions can reach users. This makes the company responsible for checking whether advertisements violate its rules.
The risks go beyond a misleading ad. Promotions for AI nudify tools can direct users to services that create fake sexual images of real people without their consent. Removing these ads after they appear does not undo the harm they may have caused.
Meta needs to provide more than figures showing how many ads it has removed. The company should explain how it screens ads before they go live, how quickly it removes violations, and how many harmful promotions are identified only after external reports. It should also explain how it stops banned advertisers from returning with new accounts.
Independent researchers should have better access to the information they need to examine Meta’s ad moderation systems. Their findings can help expose weaknesses and show how advertisers bypass existing safeguards.
The Kromix investigation shows why this oversight matters. WIRED reported on the ads, and Meta removed them after the publication contacted the company. This raises questions about why its own review systems did not block the promotions earlier.
Meta’s new AI tools may help prevent similar cases, but the company needs to show that they work. Clear data, independent scrutiny, and stronger checks before ads go live would provide better evidence of progress than removal figures alone.
Meta Must Turn Better AI Detection Into Real User Protection
Meta’s reported detection rate is important because finding child sexual exploitation content before users report it can help limit its spread. Its new AI tools for checking ad destinations and testing safety measures may also help close gaps in its existing systems.
However, Meta’s progress should be measured by how well these tools protect users, not just by the number of violations it detects.
The company reported acting on 33.2 million pieces of child sexual exploitation content in the first half of 2026. It also said it detected more than 97% of that content before users reported it. These numbers show the scale of its enforcement efforts, but they do not reveal how much harmful material its systems failed to catch.
Similarly, removing harmful ads after journalists flag them does not prove that Meta’s existing review process is working effectively.
Meta needs to show that its systems can block prohibited ads before they reach users, detect new ways of bypassing its rules, and prevent banned advertisers from returning. It should also share enough data for independent researchers and regulators to assess whether its safeguards are improving.
AI can help platforms review content faster, but it cannot solve every moderation problem on its own. Clear rules, independent oversight, transparent reporting, and action against repeat offenders are also necessary.
Meta says its AI systems are getting better at detecting abuse. The real test is whether these improvements reduce the amount of harmful content users encounter in the first place.
