Internal documents: Meta prefers to let AI assess risks in the future

Contents

Meta is planning to largely automate the risk assessment for new functions and updates to its Facebook, Instagram and WhatsApp platforms. According to internal documents obtained by US broadcaster NPR, up to 90 percent of all privacy and integrity reviews are to be carried out by AI systems instead of human reviewers in future. An exception applies to the European Union, as the legal requirements are stricter here.

Previously, all new functions and algorithm changes had to undergo a multi-stage review process in which teams of security experts assessed potential risks. For example, the reviewers analyzed whether features could violate privacy, harm minors or facilitate the spread of misinformation. Updates could only be rolled out to the billions of users once they had been approved.

Questionnaires instead of discussions

According to NPR, the new system works differently: product teams fill out a questionnaire and receive an "instant decision" from an AI system. This identifies risk areas and defines requirements that must be met before the launch. The development teams themselves are then responsible for confirming that they have implemented these requirements.

Meta wants to speed up product development and simplify decision-making processes. However, critics see this as more of a danger. Developers sometimes lack the necessary headwind. Problematic changes could be leaked more frequently.

"No data protection experts"

Experts see it as particularly problematic that the development teams are now supposed to decide on risks themselves. "Most product managers and engineers are not data protection experts," explains Zvika Krieger, Director of Responsible Innovation at Meta until 2022. Product teams are primarily measured by how quickly they launch features – This could also lead to a tendency to turn a blind eye.

Users in the European Union could be spared some of the changes. According to internal documents, decision-making for EU products and user data will remain at Meta's European headquarters in Ireland. This is due to stricter EU regulations such as the Digital Services Act, which obliges platforms to monitor harmful content more closely.

Part of Metas' realignment

The automation of risk assessments is to be part of a broader realignment at Meta, which began after the election victory of US President Donald Trump. At the beginning of the year, Meta CEO Mark Zuckerberg spoke of a cultural turning point and, among other things, relaxed the rules for combating hate speech and ended human-made fact checks. However, Meta also sees the increased use of AI automation as necessary in order to compete with TikTok, OpenAI and others.

According to the documents, the gradual introduction of AI-based assessments has been underway since April and May 2025 and was announced internally in March. The aim is to automate 90 percent of all risk assessments. Human assessors will continue to be used for complex cases for the time being, Meta explained according to NPR. The automation would only involve "low-risk decisions". The company also emphasized that it has invested billions of US dollars in user privacy.

(mki)