Voice & Audio OpenAI Opens GPT-Live-1.1 to the API and Prices Silence at 2 Cents a Minute Coding Assistants Replit Pitted 32 Claude Agents Against NumPy Tuning: 18.9x Speedup While GPT-6.2 Swarms Collapsed at Scale Security GAO Covert Tests: All Four Top Chatbots Produced Viable Attack Plans in Under 12 Hours Once Guardrails Fell Meta Meta Puts Its 2030 Net-Zero Target Under Review as AI Data Centers Push Net Emissions Up 45 Percent Business Daniel Ek Is Backing Europa Fusion, a Stealth UK Startup Led by a Former DeepMind Researcher Betting on Pulsed Power Coding Assistants Stripe Tested 14 Coding Agents on 69 Payment Scenarios: 78% of Criteria Passed, but Only 29% of Runs Met Every Requirement Security A Transcription Firm Used by Police and Prosecutors Left Interview Transcripts and Body-Cam Footage on an Open S3 Bucket News UN Killer-Robot Talks End With a 2028 Deadline and No Binding Rules. About 100 States Wanted a Treaty; the US, Russia and Israel Blocked It Voice & Audio OpenAI Opens GPT-Live-1.1 to the API and Prices Silence at 2 Cents a Minute Coding Assistants Replit Pitted 32 Claude Agents Against NumPy Tuning: 18.9x Speedup While GPT-6.2 Swarms Collapsed at Scale Security GAO Covert Tests: All Four Top Chatbots Produced Viable Attack Plans in Under 12 Hours Once Guardrails Fell Meta Meta Puts Its 2030 Net-Zero Target Under Review as AI Data Centers Push Net Emissions Up 45 Percent Business Daniel Ek Is Backing Europa Fusion, a Stealth UK Startup Led by a Former DeepMind Researcher Betting on Pulsed Power Coding Assistants Stripe Tested 14 Coding Agents on 69 Payment Scenarios: 78% of Criteria Passed, but Only 29% of Runs Met Every Requirement Security A Transcription Firm Used by Police and Prosecutors Left Interview Transcripts and Body-Cam Footage on an Open S3 Bucket News UN Killer-Robot Talks End With a 2028 Deadline and No Binding Rules. About 100 States Wanted a Treaty; the US, Russia and Israel Blocked It

Meta's AI Caught 97% of 33.2 Million Child-Exploitation Items Before Users Reported Them — Now an LLM Hunts "Signposting" Ads

Meta's AI Caught 97% of 33.2 Million Child-Exploitation Items Before Users Reported Them — Now an LLM Hunts "Signposting" Ads

Meta said it took action on 33.2 million pieces of child sexual exploitation content on Facebook and Instagram in the first half of 2026, with more than 97% detected proactively. New AI tools include an LLM system that targets "signposting" ads — benign-looking promotions that funnel users off-platform — and a red-teaming AI agent that probes Meta's own safety systems for weaknesses.

Meta announced on Wednesday that it took action on 33.2 million pieces of child sexual exploitation content across Facebook and Instagram in the first half of 2026, and said more than 97% of that content was found by its own systems before users reported it. In India alone, the company acted on 5.3 million pieces during the same period, with over 98% detected proactively.

The centerpiece of the new measures is a large language model system built to catch what Meta calls "signposting" — a tactic in which ads look entirely ordinary on the surface but are designed to steer users toward illegal content hosted elsewhere. The ads themselves may not contain illegal material; the danger lies in where they lead. Rather than analyzing only what an ad contains, Meta's system now evaluates where an ad sends users, and uses that information to block violating destinations and take action against the accounts behind them.

Meta says the shift is a response to adversaries who keep changing methods to evade detection. Alongside the signposting LLM, the company is running additional AI-driven scans designed to surface child exploitation content that earlier systems may have missed, and said it will keep adding new signals as it learns how these networks operate.

Meta is also turning AI on itself. A new "red-teaming AI agent" probes the company's own safety measures, looking for gaps that bad actors could exploit, with the goal of finding and closing abuse methods before they spread. The company said it is additionally improving systems that identify people who return to its platforms with new accounts after their previous ones were removed — the persistent re-registration problem that has long undermined enforcement.

The announcement lands under considerable legal pressure. In August, Meta reportedly reached an agreement to pay up to $18 billion to settle a child safety lawsuit involving 29 US states, and the company continues to face litigation and criticism from lawmakers over the risks its services pose to young users. This year it has rolled out a series of child-safety features, including parental controls for Meta AI, pre-teen accounts on WhatsApp and alerts when children search Instagram for self-harm content.

Two things make this announcement more than routine safety stats. First, the signposting approach — judging ads by their destinations rather than their content — is an acknowledgment that platform-boundary enforcement is now the frontier; the harmful payload rarely lives in the ad itself. Second, using an LLM and a red-teaming agent for enforcement signals that Meta sees generative models as a defensive tool at scale, not just a product surface. Detection tools in this space are not infallible — researchers have repeatedly shown AI-content detectors failing on manipulated inputs — so the 97% figure should be read as company-reported. But the direction is clear: the ad-review pipeline is becoming an AI-vs-AI contest on both sides.

Comments (0)

Log in to join the discussion

Log In

No comments yet