Meta Researcher Details How Company Manipulated Child Safety Reports

Manipulation of Research Processes

• 🔒 Research at Meta, particularly concerning child safety, underwent significant manipulation after the 2021 Frances Haugen disclosures.
• ⚖️ This manipulation involved legal teams and management monitoring every stage of research, from topic selection to data collection and final report alteration.
• 🚫 Researchers were instructed to proactively delete sensitive data and faced demands to remove or alter findings before publication.
• 🗄️ Research was increasingly siloed, limiting collaboration and the sharing of critical safety information between colleagues.

Suppression of Child Safety Research

• 📉 Meta allegedly stopped research projects into child safety because the company did not want to know the results.
• 🚧 Researchers' ability to collect information about child safety was restricted to preserve plausible deniability.
• ✏️ Research designs were altered to avoid collecting certain information, and final reports and results were modified.
• 🗑️ Researchers were required to delete data that demonstrated harm to children occurring on Meta's platforms.

Project Horton and VR User Age

• 🚀 A research project, codenamed Project Horton, aimed at understanding the age of children using virtual reality, was approved by the CTO and funded with over $1 million.
• ❌ This project was cancelled without explanation, an unusual occurrence, with Mark Zuckerberg being the only other person who could have overruled the decision.
• 📉 The cancellation is believed to be linked to the fact that improving the ability to identify underage users would require shutting down a large number of accounts, thus reducing engagement metrics.

Risks in Meta's Virtual Reality Environment

• 🧠 Children's developing ability to distinguish reality from fantasy makes them vulnerable to psychological effects from VR experiences.
• 👤 VR environments are highly social, and children often interact with strangers whose avatars appear to be the same age, making it impossible to know if they are interacting with adults.
• ⚠️ The Children's Online Privacy Protection Act requires parental consent for data collection from users under 13, but Meta lacked the ability to identify underage users on its VR platforms.
• 🗣️ In personal experience, the majority of VR users observed or interacted with were audibly under 13, a fact that was apparent to Meta leadership.

Concerns with Generative AI

• 🤖 Meta has had global initiatives targeting youth across all products for years.
• 🚫 There is no reason to believe Meta is doing more to protect children in its AI work than in its virtual reality work, especially concerning generative AI chatbots engaging in romantic or sensual conversations with kids.