Meta rolls out new AI instruments to detect advertisements that secretly result in youngster sexual abuse materials

Meta announced Wednesday that it took motion towards 33.2 million items of kid sexual exploitation content material on Facebook and Instagram within the first half of 2026. At the identical time, the corporate is rolling out new AI instruments to search out advertisements and accounts that could be used to secretly direct individuals to unlawful youngster abuse materials.

More than 97% of the content material Meta acted on was discovered by its methods earlier than customers reported it, the corporate stated. In India, Meta acted on 5.3 million items of kid sexual exploitation content material throughout the identical interval, with greater than 98% detected earlier than customers reported it.

Meta stated its newest AI instruments are geared toward individuals utilizing advertisements that look innocent to ship customers to unlawful content material exterior Meta’s platforms.

The firm has launched a brand new giant language mannequin (LLM) system to detect what it calls “signposting.” This refers to advertisements which will look regular however are suspected of directing customers to unlawful content material or different dangerous exercise elsewhere on-line. Meta says this can be a tactic it has not too long ago seen dangerous actors use as they proceed to vary their strategies to keep away from detection. While the advertisements themselves could not include unlawful materials, they will nonetheless act as a gateway to web sites internet hosting youngster sexual abuse materials.

Because of this, Meta stated it’s now the place an advert sends customers, and never simply what the advert comprises. The firm can use this data to dam web sites or different locations that break its guidelines and take motion towards the accounts behind them.

Meta can be utilizing further AI-driven scans to search out youngster exploitation content material that earlier methods could have missed. The firm stated it would proceed including new indicators because it learns extra about how these networks function.

Another new device is a “red-teaming AI agent” that assessments Meta’s personal security measures. It appears for weaknesses that dangerous actors may use to get across the firm’s protections. Meta stated this might assist it discover new strategies of abuse earlier than they change into extra widespread.

Additionally, the corporate is enhancing its methods for locating individuals who return to its platforms with new accounts after their earlier accounts have been eliminated.

The new measures come as Meta continues to face strain over the security of kids on its platforms. The firm has confronted lawsuits and criticism from lawmakers over the dangers its social media companies could pose to younger customers. In August, Meta reached an settlement to pay up to $18 billion to settle a baby security lawsuit involving 29 U.S. states, addressing issues relating to the safety of kids on-line.

Meta has launched a number of child-safety options this yr, together with parental controls for Meta AI, pre-teen accounts on WhatsApp, and alerts for folks when youngsters search Instagram for self-harm content material.

In September, WhatsApp added more controls for folks. They can restrict how youngsters use Channels, management who can see their standing updates, and resolve who can add their youngsters to teams. Parents also can select to obtain notifications about sure group exercise.