Meta is fundamentally reshaping its approach to product safety and privacy, intending for an artificial intelligence system to manage up to 90% of risk evaluations for updates across its widely used applications, including Instagram and WhatsApp.
This significant operational change, based on internal documents revealed by an NPR investigation, aims to accelerate product development. While users might see new features rolled out more rapidly, the move has ignited concerns about the depth of safety and privacy scrutiny.
The new AI-driven process, which NPR reports has been ramping up through April and May, involves product teams completing a questionnaire and receiving an “instant decision” from the AI, which identifies risks and outlines pre-launch requirements. This shift occurs under the shadow of a 2012 agreement with the Federal Trade Commission that mandates thorough privacy reviews—duties historically performed by human staff.
An unnamed former Meta executive expressed unease to NPR, stating, “Insofar as this process functionally means more stuff launching faster, with less rigorous scrutiny and opposition, it means you’re creating higher risks,” and warned that “Negative externalities of product changes are less likely to be prevented before they start causing problems in the world.” In response, Meta has asserted that automation will be confined to “low-risk decisions,” with “human expertise” reserved for “novel and complex issues.”
The Shift To AI-Driven Oversight
This automation of risk assessment is a key part of Meta’s broader, aggressive strategy to embed AI across its operations, a direction increasingly evident since early 2025. The company’s substantial commitment includes a planned $65 billion investment in AI this year.
This financial dedication is coupled with significant corporate restructuring, which involves Meta doubling down on machine learning by planning to hire hundreds of AI engineers while cutting 5% of its overall workforce.
Michel Protti, Meta’s chief privacy officer for product, announced in a March internal post that the company is “empowering product teams” and “evolving Meta’s risk management processes.”
The goal, Protti explained, is to “simplify decision-making” by automating risk reviews in the vast majority of cases, as per the NPR investigation. This internal push for speed and efficiency through AI also extends to content moderation.
Meta’s latest quarterly integrity report, cited by NPR, claims that Large Language Models (LLMs) are “operating beyond that of human performance for select policy areas” and are used to screen posts “highly confident” not to violate rules.
However, internal documents reviewed by NPR suggest Meta is considering automating reviews even for sensitive areas like AI safety, youth risk, and ‘integrity’—which encompasses violent content and misinformation—despite public statements focusing on low-risk automation.
Balancing Innovation, Safety, and Regulatory Scrutiny
The drive towards AI-led oversight is influenced by intense competition in the tech landscape. The performance of rival AI models, such as DeepSeek’s R1, has reportedly created a sense of urgency within Meta.
An engineer previously described a “mad scramble trying to match that efficiency.” This competitive environment is a significant factor in Meta’s strategic decisions, including leadership changes like Loredana Crisan, formerly head of Messenger, now overseeing the company’s generative AI division.
Meta’s approach to AI governance has been under development for some time. In Februar, the company introduced its Frontier AI Framework, a system designed to categorize AI into “high-risk” and “critical-risk” groups.
At its launch, Meta stated its intent: “Through this framework, we will prioritize mitigating the risk of catastrophic harm while still enabling progress and innovation.”
This initiative was, in part, a response to past incidents, such as the misuse of its LLaMA models, and the increasing pressure from regulations like the European Union’s Digital Services Act (DSA).
Zvika Krieger, a former director at Meta, commented to NewsBytes that while automation can streamline reviews, “if you push that too far, inevitably the quality of review and the outcomes are going to suffer.” Interestingly, an internal Meta announcement indicated that decision-making and oversight for products and user data in the EU will remain with Meta’s European headquarters in Ireland, potentially insulating EU users from some of these changes, according to the NPR report.
Broader AI Integration and Partnerships
Meta’s AI ambitions extend beyond internal processes and consumer-facing products. The company updated its ‘acceptable use’ policy in November 2024, permitting US military firms to utilize its large language AI models. Companies including Lockheed Martin, Booz Allen Hamilton, Palantir Technologies, and Anduril Industries can now leverage Meta’s AI tools.
This includes a partnership with Anduril Industries to develop advanced military equipment, such as AI-powered helmets with VR and AR capabilities. Meanwhile, Meta’s Q1 2025 Community Standards Enforcement Report highlighted a roughly 50% reduction in enforcement mistakes in the US compared to Q4 2024, an improvement attributed to focusing on high-severity violations and enhancing accuracy through system audits.
