Home · Technology · Oct 7 archive
Meta Introduces New AI Tools to Detect Ads Linked to Child Exploitation
Confirmed
In Short: Meta announced Wednesday that it took action against 33.2 million pieces of child sexual exploitation content on Facebook and Instagram in the first half of 2026, with more than 97% of the content detected by its systems before users reported it.
Meta announced Wednesday that it took action against 33.2 million pieces of child sexual exploitation content on Facebook and Instagram in the first half of 2026, with more than 97% of the content detected by its systems before users reported it.
In India, Meta acted on 5.3 million pieces of child sexual exploitation content during the same period, with more than 98% detected before users reported it.
The company has introduced a new large language model (LLM) system to detect what it calls “signposting.” This refers to ads that may look normal but are suspected of directing users to illegal content or other harmful activity elsewhere online.
According to researchers at the nonprofit Tech Transparency Project, more than 250 additional ads containing child sexual abuse material (CSAM) have run across Meta’s social media platforms since the start of August, collectively shown to more than 29,000 accounts in countries across the European Union, plus around 7,000 accounts in the United Kingdom.
The ads were animated to depict disturbing scenes involving real-life minors, including a 14-year-old Instagram influencer and a preteen posing in a pink athletic outfit.
Meta confirmed it will directly report child sexual abuse material to Indian law enforcement, enhancing child safety measures.
Despite Meta’s claims of enhanced detection tools, researchers found that the company failed to detect 332 ads containing CSAM this year.
Meta’s new detection tools were supposed to block such content, but the ads continued to appear after researchers alerted the company.
Pham, a researcher, told Wired, “Every time the platform or investigators would build a new detection model or a new way to have some type of enforcement, they quickly — within a day, maybe five days — built a new way to avoid detection.”
What this adds
Meta's new AI tools are aimed at preventing ads that secretly lead to illegal content, but the company has faced criticism for its failure to detect and remove such ads promptly.
The ongoing issue highlights the challenges in effectively monitoring and preventing the spread of child sexual abuse material on social media platforms.
What's confirmed
- Meta announced Wednesday that it took action against 33.2 million pieces of child sexual exploitation content on Facebook and Instagram in the first half of 2026, with more than 97% of the content detected by its systems before users reported it.
- In India, Meta acted on 5.3 million pieces of child sexual exploitation content during the same period, with more than 98% detected before users reported it.
- The company has introduced a new large language model (LLM) system to detect what it calls “signposting.” This refers to ads that may look normal but are suspected of directing users to illegal content or other harmful activity elsewhere online.
- According to researchers at the nonprofit Tech Transparency Project, more than 250 additional ads containing child sexual abuse material (CSAM) have run across Meta’s social media platforms since the start of August, collectively shown to more than 29,000 accounts in countries across the European Union, plus around 7,000 accounts in the United Kingdom.
- The ads were animated to depict disturbing scenes involving real-life minors, including a 14-year-old Instagram influencer and a preteen posing in a pink athletic outfit.
- Meta confirmed it will directly report child sexual abuse material to Indian law enforcement, enhancing child safety measures.
- Despite Meta’s claims of enhanced detection tools, researchers found that the company failed to detect 332 ads containing CSAM this year.
- Meta’s new detection tools were supposed to block such content, but the ads continued to appear after researchers alerted the company.
What's still developing
- Get your pass and bring someone with you at 50% off.
- At the time, the company said that most of the abusive ads had been published prior to it rolling out new AI tools designed to “better detect and block” harmful ads being uploaded.
- Overall, the researchers discovered that Meta has published more than 350 of the abusive video ads since the end of last year, with many of them linking to harmful harmful “ nudification ” apps linked to Chinese developers.
- “You psycho, why are you abusing the child,” wrote one user on a video of a woman slapping a boy in the face with a belt, leaving a bloody mark around his eye.
- Just this week, Wired reported that the company has accepted money from advertisers to run hundreds of promos for “nudify” apps that contained AI-generated sexual imagery of children, including numerous real-life minors whose photos were pilfered from social media and stock images.
- Can’t-miss innovations from the bleeding edge of science and tech Content warning: this story contains descriptions of disturbing AI-generated videos that depict violent child abuse.
- Michelle Kuppersmith, executive director of Campaign for Accountability, which runs TTP, noted that the report comes after a historic $18 billion settlement forced Meta to agree to make changes to platforms to enhance child safety.
- Other ads animated a photo of a 14-year-old Instagram influencer “showing off her new sports club uniform” into “a video of her performing oral sex.” A third “preteen” victim “posing in a pink athletic outfit with pigtails” in a series of stock photos was morphed into a video where she looks frightened as she’s molested by an adult male, TTP reported.
- For 43 of these, TTP got a reply that said, "We've taken a look and found that this ad doesn't go against our Advertising Standards."
- In July, after BBC's investigation, Meta had put out a detailed post about how it fights child exploitation on its apps When the BBC reached out to Meta for a comment, they said, "we've built strong defenses, including systems that block violating ads as well as ongoing monitoring in case we initially miss something: that's why many of the ads flagged to us had already been removed, most had fewer than 200 impressions, and the total ad spend was under $5,000".
- The document was also signed by executives from Google, Anthropic, Meta and more An agreement signed by President Donald Trump and some of the biggest names in artificial intelligence had a very clear typo: the misspelling of the United States.
- Of the 50,000 voters who participated, “Supreme Intelligence” racked up 65% of votes.