Ad
Image credit: SofiasHunkina via Envato
Meta is transitioning the majority of its internal safety and privacy reviews to artificial intelligence, replacing a system that has traditionally relied heavily on human judgment.
Internal documents first obtained by NPR reveal that up to 90% of Meta’s risk assessments are expected to be automated. Previously, specialized teams evaluated how updates could impact user privacy, harm minors, or facilitate the spread of misinformation. Under the new system, the responsibility for these assessments will largely be transferred to AI technologies.
Meta is the parent company of Facebook, Instagram, WhatsApp, and Threads.
AI to decide on product risks
Under this new framework, product teams will fill out a questionnaire detailing updates, after which an AI system will provide an immediate decision, identifying potential risks and establishing necessary conditions for the project. Human oversight will only be required in select cases, such as when projects introduce novel risks or when a team especially requests it. A slide from Meta’s internal presentation describes this process as one where teams will “receive an ‘instant decision’” based on AI evaluation.
This shift allows developers to release features much faster. But experts, including former Meta insiders, worry that speed is coming at the cost of caution.
“Insofar as this process functionally means more stuff launching faster, with less rigorous scrutiny and opposition, it means you’re creating higher risks,” a former Meta executive told NPR on the condition of anonymity.
Meta, in a statement, said the new process is designed to “streamline decision-making” and that “human expertise” will still be used for “novel and complex issues.” The company insisted that only “low-risk decisions” are being automated; however, internal documents obtained by NPR reveal that more sensitive areas — such as AI safety, youth risk, and content integrity (including violent or false content) — are also slated for automations.
More must-read AI coverage
Critics say it could backfire
Some within and outside Meta caution that over-reliance on AI for risk assessments could be shortsighted. Another former Meta employee, who spoke to NPR under anonymity, said: “This almost seems self-defeating. Every time they launch a new product, there is so much scrutiny on it — and that scrutiny regularly finds issues the company should have taken more seriously.”
Katie Harbath, former public policy director at Facebook and now chief executive officer of Anchor Change, offered a more balanced view.
“If you want to move quickly and have high quality you’re going to need to incorporate more AI, because humans can only do so much in a period of time,” she told NPR. She emphasized that “those systems also need to have checks and balances from humans.”
Regulatory pressure and European exceptions
Since 2012, Meta has been under a Federal Trade Commission (FTC) agreement that requires it to conduct privacy reviews for product updates. This oversight followed a settlement concerning the company’s handling of user data.
In response to these obligations, Meta said it has “invested over $8 billion in our privacy program” and continues to refine its processes. “As risks evolve and our program matures, we enhance our processes to better identify risks, streamline decision-making, and improve people’s experience,” a company spokesperson told TechCrunch.
Interestingly, European Union users may not face the same level of automation. Internal communications indicate that decision-making for EU-related products will still be managed by Meta’s headquarters in Ireland, in part due to the Digital Services Act, which imposes stricter rules on content and data protection.
The shift toward automation aligns with other recent policy changes at Meta, including the phase-out of its fact-checking program and relaxation of its hate speech policies.
In its Q1 2025 Integrity Report, Meta highlighted that its AI systems are already outperforming humans in some policy areas. “This frees up capacity for our reviewers allowing them to prioritize their expertise on content that’s more likely to violate,” the company wrote.
Ad
