On October 7, 2026, Meta announced the launch of new artificial intelligence (AI) tools designed to detect ads that lead to material related to the sexual exploitation of children. Both the company itself and TechCrunch reported the news on the same day — TechCrunch covered the details of the new tools, while the company blog highlighted the child-safety measures taken in India. According to TechCrunch, the new LLM specifically targets the hidden redirects behind ads. The announcement covers two fronts: a large language model (LLM) that tracks where an ad leads, and a separate AI agent that tests its own safeguards. According to the company, the new tools are aimed at strengthening ad oversight on Facebook and Instagram. Both announcements were made on October 7, 2026.
Ads Are Checked for Where They Lead
The LLM launched by the company specializes in detecting ads known as "signposting." "Signposting" refers to a sign or indicator pointing the way: such ads can appear harmless, but their purpose is to direct users to illegal child sexual abuse material hosted outside the platform. The system was built specifically to find such hidden redirects — to detect them, Meta relies on a large language model whose job is to identify signs of concealed redirection.
The key difference in the new approach is that Meta now tracks not just what an ad shows, but where it takes the user. If an ad leads to an off-platform destination where a violation has been detected, the company blocks the paths leading to that destination and takes action against the accounts behind such ads. In other words, accountability does not stop at the ad copy — it reaches the account level. The measures are applied in two stages: first the ad leading to that destination is blocked, then action is taken against the account that placed the ad.
Meta explained the mechanism as follows:
"Meta now scans where an ad leads, not just what it shows, so it can block violating off-platform destinations and take action against the accounts behind them." — Meta (TechCrunch, October 7, 2026)
This means oversight no longer stops at ad copy and imagery — the check extends to the external page that opens after the ad is clicked.
A Separate AI Agent That Tests the Safeguards
Meta also introduced a "red-teaming" AI agent that actively tests its own safety defenses. Its task is to deliberately probe the company's protection systems and find vulnerabilities that bad actors could exploit.
The agent's work is aimed at identifying gaps in the defensive line before new abuse methods spread widely. Its scrutiny is directed not at external threats, but at the protection systems themselves — the agent hunts for weaknesses in the defenses. Put differently, the system does not wait for violators to find new routes — it proactively searches for their likely paths and exposes weak points at an early stage.
The company is thus moving on two fronts at once: on one hand, an LLM that finds "signposting" ads has been launched; on the other, an agent that continuously tests the safeguards themselves has been introduced. The first tool catches an existing threat, the second checks whether the defenses are ready for future tests.
First-Half Numbers
According to the report published by Meta, action was taken on 33.2 million pieces of content related to child sexual exploitation across Facebook and Instagram in the first half of 2026. More than 97% of them were detected by the company's own systems before users complained — showing that the main burden of detection rests on automated systems.
India-specific figures were reported separately: from January to June 2026, action was taken on 5.3 million such pieces of content in the country, and more than 98% of them were found and removed before anyone reported them. The global 33.2 million and the India-only 5.3 million both refer to the first six months of 2026 — the January–June period — and in both cases the majority share was detected by automated systems before user reports.
Direct Reporting Process in India
Meta said it will now send child-safety cases in India directly to the country's National Cybercrime Reporting Portal. The portal is managed by the Indian Cyber Crime Coordination Centre (I4C). According to the company, cases will be delivered directly to the national portal under this process.
The tool details were covered in the TechCrunch publication; the measures in India were reported by Meta's official blog.