Facebook and Instagram users are being targeted by child sexual exploitation content, which Meta removed on 7th October from 33.2 million instances. The company is also launching new AI capabilities to search for ads and accounts that could be used to stealthily direct individuals to illegal child pornography.

The company said it had acted on over 97% of the content discovered through its systems prior to users reporting it. During the same timeframe, Meta took action against 5.3 million reports of child sexual exploitation on its platforms in India, a rate surpassing the 98.4% threshold of pre-user-reported content detection.

Meta claims its new AI capabilities are designed to target users of seemingly benign ads directing them to illegal content on platforms outside of Meta.

The firm has launched a brand new Large Language Model (LLM) system to recognise what it calls “signposting.” These are ads that might appear normal but are believed to lead to illegal content or other bad behaviour on the Internet. Meta states that it has recently observed the same strategy being employed by bad actors as they continually evolve their strategies to evade detection. The ads may not include illegal content, but they can be a gateway to sites containing child sexual abuse material.

For this reason, Meta is now examining the destination of an ad, rather than its content. The company can use it to block sites or other destinations that violate its rules, and can take action against the accounts responsible for doing so.

Children’s content is also being scanned using more AI tools to identify such material, which may have been missed by the previous tools. The company said it will keep adding signals as it gathers more information about the operation of these networks.

A “red-teaming AI agent” is another new feature that is meant to test Meta’s own safety features. It scans for vulnerabilities that could be exploited by hackers to circumvent the business’s measures. That means this might assist the enterprise in identifying new ways to be abused before they become more widespread, Meta said.

Besides, the business is enhancing its processes to quickly identify and locate individuals who resume their use of its platforms with new accounts when their original accounts are removed.

The changes are part of a series of actions taken by Meta as it continues to receive pressure about its platforms’ safety for children. The company has been sued and criticised by politicians for the potential dangers its social networking services could bring to young users. Meta also agreed to pay up to $18 billion to settle a child safety lawsuit brought by 29 U.S. states in August, due to worries about how children are safeguarded online.

This year, there are a number of new features we’ve seen that help keep kids safe on Meta, such as parental controls for Meta AI, preteen accounts on WhatsApp, and alerts to parents when they’re looking at self-harm content on Instagram.

More controls for parents were introduced in September on WhatsApp. They can restrict teens’ use of Channels, regulate who can view their status updates, and determine who can add their kids to groups. Parents may also opt to be informed of some group activity.