What happened
Meta announced new AI tools to detect ads that secretly link to child abuse material. According to TechCrunch, the company is using a new large language model to spot "signposting" — ads that appear harmless but direct users to illegal content off-platform. The system analyses an ad's destination, not just its content.
This comes as Meta reported it took action on 33.2 million pieces of child sexual exploitation content in the first half of 2026. The company also deployed a "red-teaming AI agent" to proactively find weaknesses in its own safety systems.
How the room's reading it
The move is widely seen as a direct response to sustained legal and regulatory pressure. Trust and safety practitioners note that this isn't happening in a vacuum — Meta has recently faced significant lawsuits over child safety on its platforms. The technical shift from analysing ad content to scanning ad destinations is the main point of discussion for developers in the space.
The consensus is that this raises the bar for platform moderation. It signals a move from policing on-platform content to assessing off-platform risk. The use of a red-teaming AI is also being framed as an emerging best practice, where platforms use adversarial AI to harden their own defences before bad actors find the exploits.
Sailfish's take
We see this as a clear signal for anyone building a platform with user-generated content or third-party ads. Your responsibility no longer ends at your site's border. The core problem Meta is chasing here — "signposting" — isn't unique to them. Any service that allows outbound links has to consider where it's sending its users.
We've shipped enough products with user content to know that simple keyword filtering is a solved problem. The real challenge is understanding intent. Meta's move confirms that the next frontier for trust and safety is analysing user behaviour and destinations, not just static content. If you're building a platform today, this is the week to start auditing your outbound link policies.