Meta AI policy leak revealed chatbots permitted flirty conversations with minors

Amid rising concerns about emotionally manipulative AI chatbots like ChatGPT, Reuters reports that Meta permitted its AI personas to flirt with minors, spread misinformation, and produce discriminatory responses targeting marginalized groups.

An internal Meta document obtained by Reuters revealed policies permitting AI chatbots to initiate "romantic or suggestive dialogues with minors."

Meta verified the document's authenticity to Reuters, confirming it contained governance rules for Meta AI and its messenger chatbots. These standards reportedly received approval from Meta's legal, policy, engineering teams, and chief ethics officer.

The report coincides with another Reuters investigation detailing how Meta's flirtatious AI persona deceived a retiree into believing it was human, ultimately contributing to his fatal accident after luring him to a New York address.

While previous reports highlighted Meta's sexually suggestive AI interactions with minors, Reuters' investigation raises deeper concerns about Meta's AI companion strategy potentially exploiting what CEO Mark Zuckerberg termed the "loneliness epidemic."

The 200-page "GenAI: Content Risk Standards" document included sample exchanges demonstrating approved responses to inappropriate prompts. One example showed an AI endorsing romantic exchanges with teenagers while prohibiting explicit sexual roleplay.

"Our policies prohibit inappropriate interactions with minors," Meta spokesperson Andy Stone told TechCrunch. "Incorrect annotations were erroneously included in the document and have since been corrected."

Stone clarified that Meta has removed these guidelines and now prevents romantic AI exchanges with minors, though still permits interactions with users aged 13+.

Sarah Gardner of Heat Initiative remained skeptical, demanding transparency: "If Meta has truly rectified this, they must publicly release updated guidelines showing exactly how their AI interacts with children."

Harmful content policies under scrutiny

Reuters found that while prohibiting hate speech, Meta's policies permitted AI to generate demeaning statements about protected groups.

One alarming example showed Meta's AI constructing racist pseudoscientific arguments when prompted. This revelation follows Meta's hiring of conservative activist Robby Starbuck to address perceived political biases in Meta AI.

The standards also permitted fabricated statements when labeled as untrue, and allowed potentially harmful financial/medical advice prefaced with disclaimers.

Regarding celebrity image generation, policies prohibited explicit requests but contained bizarre loopholes - permitting modified nude images replacing hands with objects like "an enormous fish." Meta's Stone insisted "the guidelines never allowed nude images."

Violence policies showed similar contradictions, allowing depictions of child brawls and elder abuse while prohibiting gore.

Ethical concerns escalate

Critics accuse Meta of employing dark patterns that manipulate users, particularly teens. Persistent visibility of "like" counts continues despite internal research showing psychological harm.

Whistleblower Sarah Wynn-Williams revealed Meta once profiled teens' emotional vulnerabilities for targeted advertising.

Meta actively opposed the Kids Online Safety Act, legislation addressing social media's mental health impacts. Though initially defeated in 2024, senators reintroduced the bill this year.

Recent TechCrunch reporting indicates Meta is developing proactive AI chatbots that initiate unprompted conversations, mirroring features from startups facing lawsuits over harmful interactions with minors.

With 72% of teens using AI companions, experts increasingly warn about adolescents forming unhealthy attachments to chatbots at the expense of real-world relationships.

Tech leaders confirmed for Disrupt 2025

Netflix, ElevenLabs, Wayve, Sequoia Capital, and Elad Gil headline TechCrunch Disrupt 2025. Gain actionable insights from tech's most influential voices during our 20th anniversary event. Secure your tickets now with savings up to $600 before prices increase.

Tech leaders confirmed for Disrupt 2025

Netflix, ElevenLabs, Wayve, and Sequoia Capital join Disrupt 2025's speaker lineup. Learn growth strategies from industry leaders at our milestone 20th anniversary event. Register today to save up to $675 on passes.

San Francisco | October 27-29, 2025 REGISTER NOW