Oversight Board

Many congratulations to Board Member Nighat Dad for being named in the Chief in AI Global 100 list!

The Board has selected a case involving an Instagram user’s post sharing her personal experience with prescription drugs used in pregnancy and childbirth. The post was taken down under Meta’s Restricted Goods and Services policy and then restored with an age-rating after the Board took on the case following a user appeal. Visit our website to find out more about the case and how you can contribute valuable perspectives before 23:59 Pacific Standard Time (PST) on Tuesday 11 August. 🔗 https://lnkd.in/gbSc7WGb 🔗 #Meta #OversightBoard #pregnancy #pregnant

The Board has selected a case to examine the global issue of manipulated media being used to facilitate online scams and the effectiveness of Meta’s policies to address the issue. The specific case the Board is considering involves an AI-generated video of a monk who appears to be part of the Walk for Peace march asking for support, with a link to Patreon. Visit our website to find out more about the case and how you can contribute valuable perspectives before 23:59 Pacific Standard Time (PST) on Thursday 6 August. 🔗 https://lnkd.in/g9i4wyCi 🔗

The Board’s first-ever evaluation of AI models and political speech assessed output from 10 major LLMs and found they are more than twice as likely to refuse requests criticizing repressive regimes compared to speech-permissive ones. You can read the full report on our website. 🔗 https://lnkd.in/ev-hbZ2i 🔗

The Board has selected a case involving an Instagram user’s post sharing her personal experience with prescription drugs used in pregnancy and childbirth. The post was taken down under Meta’s Restricted Goods and Services policy and then restored with an age-rating after the Board took on the case following a user appeal. Visit our website to find out more about the case: https://lnkd.in/dXzskK2e The Board would appreciate public comments that address: The impact of putting page and recommendation restrictions on content of this nature, including implications for access to health information. The prevalence and impact of overenforcement against reproductive and maternal health content on Meta's platforms. How Meta should evaluate content referencing high-risk or controlled substances, such as fentanyl, when described in a documented, clinical context such as childbirth. How best to ensure due process and fairness to users whose accounts are penalized. If you or your organization feel you can contribute valuable perspectives that can help with reaching a decision, you can submit your contributions via our website before 23:59 Pacific Standard Time (PST) on Tuesday 11 August. #Meta #OversightBoard #pregnancy #pregnant

Board Member Nighat Dad led a roundtable at TrustCon in San Francisco about child safety regulation online. The roundtable was co-hosted with Alice Goguen Hunsberger, Head of Trust & Safety for Musubi, together with Hilary R. from Global Network Initiative (GNI) as guest speaker. The conversation focused on the wider debate about age-based social media bans and how we can build trust for both adults and children online, rather than create policies that treat them as entirely separate audiences. Guests in attendance included representatives of platforms such as TikTok, Anthropic and Meta and organizations such as Roblox, The Pokémon Company International and UNICEF. Thank you to everyone who joined the conversation and pushed the thinking forward. Trust & Safety Professional Association

The Board’s first-ever evaluation of AI models and political speech assessed output from 10 major LLMs and found they are more than twice as likely to refuse requests criticizing repressive regimes compared to speech-permissive ones. Watch as Co-Chair khaled Mansour and Board Member Suzanne Nossel give a rundown of the key findings. 🔗 You can read the full report on our website: https://lnkd.in/ev-hbZ2i 🔗 #OversightBoard #Meta #AI #LLM #Report

The Board has selected a case to examine the global issue of manipulated media being used to facilitate online scams and the effectiveness of Meta’s policies to address the issue. The specific case the Board is considering involves an AI-generated video of a monk who appears to be part of the Walk for Peace march asking for support, with a link to Patreon. Visit our website to find out more about the case. 🔗 https://lnkd.in/etpvqvGd The Board would appreciate public comments that address: 👉 The effectiveness of Meta’s scam policies in addressing potentially misleading donation campaigns. 👉 The prevalence of scams linked to religious, social or charitable causes across platforms, how they are typically facilitated and their impact on users. 👉 The use of manipulated media to facilitate online scams. 👉 Best practices and recommendations for moderating scams facilitated by AI-generated content. 🫵 If you or your organization feel you can contribute valuable perspectives that can help with reaching a decision, you can submit your contributions via our website before 23:59 Pacific Standard Time (PST) on Thursday 6 August.

Could LLMs be stifling political speech? Board Member Nicolas Suzor has written a blog following our report assessing how some of the world’s most-used models from Anthropic, DeepSeek, Google, Meta and OpenAI are significantly less likely to criticize political regimes that restrict free expression. https://lnkd.in/desMi-By #OversightBoard #Meta #AI #Report #LLMs #FreeSpeech

I thoroughly enjoyed participating in last week's discussion hosted by Oversight Board: "An Oversight Body for AI: Lessons in Global Content Moderation." We explored what AI can learn from years of social media content moderation, how international human rights principles should be applied to these emerging technologies, and how AI can help identify when and where human moderation is most needed. My thoughts from the discussion: 1. Human oversight remains essential. AI can improve efficiency, but it cannot replace human judgment in complex or high-impact decisions. Transparency builds trust, and meaningful appeals and accountability are critical. Bias must be continuously monitored, and a risk-based approach to moderation is key. 2. Key questions remain for AI-enabled content moderation. These include clearly defining appropriate use cases and purposes, distinguishing between moderation in public versus private channels, and determining how to address the propagation of strictly illegal content while respecting fundamental rights. As AI changes how, what, and the speed and scale at which content is produced, organizations need to also think beyond content moderation as a solution and examine more deeply the structure and design of systems and platforms. 3. AI does not shift accountability. Given the scope and scale that AI enables, for many organizations, the greater challenge may be the propagation—as opposed to the suppression—of harmful content, though both raise important questions about the impact on fundamental rights. As decisions become increasingly automated, or even partially automated, important questions arise: What level of internal risk assessment has been conducted? How are potential harms evaluated? And what costs and consequences are organizations prepared to accept? The Meta Oversight Board continues to provide thoughtful, detailed, and case-specific guidance on some of the most challenging questions surrounding content moderation. I was grateful for the opportunity to contribute to the conversation and thank the Oversight Board for inviting me to join the panel.