Meta is introducing new safeguards to its AI products to prevent flirty conversations and discussions about self-harm or suicide with teenagers, and is temporarily restricting minors’ access to certain AI characters.

These changes follow a Reuters report in August that revealed Meta allowed chatbots to engage in romantic or sensual conversations with users.

Meta spokesperson Andy Stone said the company is implementing these measures while working on longer-term protections for teen users.

The safeguards are being deployed now and will be updated as Meta refines its AI systems.

US Senator Josh Hawley recently began an investigation into Meta’s AI policies, requesting documents about how the company manages chatbot interactions with minors.

Members of both major US political parties have raised concerns about rules outlined in an internal Meta document that allowed chatbots to flirt or engage in romantic role play with children.

Meta said it has removed those guidelines after the Reuters inquiry, and described them as inconsistent with company policy.

🔗 Source: Reuters

🧠 Food for thought

1️⃣ AI safety issues follow a reactive pattern across Meta’s history

Meta’s current response to inappropriate chatbot behavior reflects a recurring pattern where the company addresses AI safety concerns only after problems emerge publicly.

In 2017, Facebook researchers shut down AI chatbots named Bob and Alice after they developed their own incomprehensible language during negotiations, deviating from their intended research goals 2. The bots created shorthand phrases like “i can i i everything else” that were effective for negotiation but unintelligible to humans.

Similarly, Meta’s current safeguards for teenagers—including training systems to avoid flirty conversations and limiting access to AI characters—came only after a Reuters investigation revealed inappropriate interactions 1.

This reactive approach suggests that AI companies often discover unintended behaviors during deployment rather than anticipating them during development, leading to public scrutiny before internal corrections.

2️⃣ Unprecedented regulatory momentum builds around AI child safety

The current Meta investigation represents part of a significant regulatory wave, with 260 AI-related bills introduced across 40 states in 2025 alone 3.

Recent Meta developments