Ethics Report: Meta Superintelligence Labs
Rubric: Organisation v4 · Reviewed 4/1/2026
Minimal ethical infrastructure
Safety & Harm Reduction
8/25Dedicated safety / responsible-use policy
Publishes a dedicated safety/responsible-use policy that is publicly accessible.
Evidence
Meta Superintelligence Labs has a 'Meta Code of Conduct' that mandates 'building responsibly' and integrates safety protocols directly into reinforcement learning and red-teaming processes. However, no dedicated safety/responsible-use policy page with specific, enforceable terms and defined prohibited uses is described in publicly available documentation. The 'Responsible Use Guide' is mentioned as outlining a multi-stage process for LLM product development, but this appears to be a general development guide rather than a comprehensive safety policy. This constitutes a generic safety values statement rather than a full dedicated policy.
Sources
- •Article Content - Safety & Ethics section
Public bug-bounty or red-team program
Operates or funds a public bug-bounty / red-team program. Internal-only programs score 0.
Evidence
Meta Superintelligence Labs employs FERRET (Framework for Expansion Reliant Red Teaming), an automated multi-modal framework designed to stress-test models through adversarial conversations. The article documents that Meta uses red-teaming processes and mentions this automated approach identifies vulnerabilities. However, no evidence is provided that this is a publicly documented program with published results from at least one completed round. The red-teaming program exists but results publication is not evidenced.
Sources
- •Article Content - Safety & Ethics section
Published safety evaluation within last 24 months
Published safety evaluation/audit/model card within last 24 months with quantitative benchmarks on harmful outputs (bias, toxicity, hallucination, etc.).
Evidence
The article mentions that Meta Superintelligence Labs maintains a high volume of peer-reviewed publications and presents findings at academic conferences such as NeurIPS, ICML, CVPR, and ICLR. However, no specific published safety evaluation within the last 24 months is described. The reference to 'Responsible Use Guide' suggests some documentation exists, but comprehensive quantitative safety benchmarks across multiple harm categories or third-party audits are not detailed. This indicates limited safety documentation.
Sources
- •Article Content - Research & Development section
Documented content-filtering / guardrails
Documents content-filtering/guardrails on production endpoints with user-facing documentation.
Evidence
Meta provides a 'Responsible Use Guide' that outlines a multi-stage process for LLM product development and deploys 'Llama Guard,' a specialized safety model that filters both input prompts and model outputs based on a predefined taxonomy of risks. The article states that guardrails are designed to mitigate risks at both system and input levels. However, detailed documentation explaining what is filtered, why, and how users can report false positives is not provided in the article.
Sources
- •Article Content - Safety & Ethics section
Documented incident-response process
Documented incident-response process for safety failures with a reporting mechanism. Generic "contact us" alone = 0.
Evidence
No documented incident-response process with a reporting mechanism, SLAs, or response timelines is described in the article. While Meta has safety and moderation teams, no public process for reporting safety incidents or security issues is detailed. Generic 'contact us' mechanisms are not mentioned either. The article does not establish evidence of a dedicated incident-response process.
Sources
- •Article Content - Safety & Ethics section
Transparency & Trust
9/25Training data provenance disclosure
Publishes training data provenance disclosures identifying sources/types/datasets. "Publicly available data" alone = 0.
Evidence
The article states that Meta uses an 'open-weights' philosophy and provides access to model weights for the Llama series. However, specific training data provenance is not detailed. The article mentions that Meta trains models on 'publicly available data' and notes that Meta uses public posts from Instagram and Facebook to train AI systems, but does not provide specific dataset names, sources, or meaningful composition/curation detail. This constitutes general categories rather than specific datasets.
Sources
- •Article Content - Lead and Products & Services sections
Meaningful technical documentation for flagship model(s)
Publishes meaningful technical documentation (system card, tech report, research paper) for flagship model(s) regardless of whether weights are released.
Evidence
The article extensively documents Meta's technical approach to the Llama series, including that Llama 3.1 features a 405B parameter model achieving 88.5% on HellaSwag benchmark, uses a 128,000-token context window, and employs Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning from AI Feedback (RLAIF) for alignment. The article covers architecture details, scale, training approach (including alignment methods), and discusses limitations related to the trade-off between open-weights accessibility and safety controls. This substantive technical documentation meets the criteria for a score of 5.
Sources
- •Article Content - Products & Services and Safety & Ethics sections
Transparency report (takedowns, government requests, etc.)
Publishes a transparency report covering takedowns, government requests, enforcement stats, and/or safety incidents.
Evidence
No transparency report covering government requests, takedowns, or content moderation decisions is described in the article. While the article mentions the Meta Oversight Board reviewing content moderation and policy decisions and issuing over 300 policy recommendations, this is not a transparency report from Meta Superintelligence Labs disclosing government requests or takedown statistics. No comprehensive or recent transparency report meeting the criteria is evidenced.
Sources
- •Article Content - Safety & Ethics section
ToS training data use disclosure with opt-out
ToS explicitly states whether user inputs/outputs are used for training, with opt-out mechanism if applicable.
Evidence
The article states that Meta admits it uses public posts from Instagram and Facebook to train its AI systems and notes this has drawn sharp criticism from privacy advocates for bypassing user intent without an explicit opt-in mechanism. The ToS or training data use disclosure is either vague or absent regarding user opt-out mechanisms. The article indicates no opt-out or opt-in mechanism is currently provided to users for training data use.
Sources
- •Article Content - Reception & Controversies section
Creator/artist content provenance disclosure
Discloses training data provenance specifically for creator/artist content (copyrighted or artist-created works).
Evidence
The article acknowledges that Meta Superintelligence Labs' models were trained on large-scale datasets containing copyrighted works, and that the Authors Guild and prominent novelists filed a class-action lawsuit regarding this. The article also mentions that artist organizations and intellectual property experts argue the models were trained without adequate compensation or opt-out mechanisms. However, no specific disclosure naming content types, sources, or licensing arrangements for creative works is provided. This constitutes a general acknowledgment without specific details.
Sources
- •Article Content - Reception & Controversies and Societal Impact sections
Human & Creator Impact
2/25Artist/creator opt-out or removal mechanism
Documented artist/creator opt-out or removal mechanism. "We respect copyright" alone = 0.
Evidence
The article does not describe any artist/creator opt-out or removal mechanism. While the Authors Guild lawsuit and criticism from artist organizations are mentioned, indicating awareness of the issue, no documented opt-out form, process, email removal mechanism, or evidence of honoring removal requests is provided. The article states that critics argue models were trained without 'opt-out mechanisms,' indicating no such mechanism exists.
Sources
- •Article Content - Reception & Controversies section
Public licensing or revenue-sharing with creators
Public licensing agreements or revenue-sharing partnerships with creators/publishers/media organizations.
Evidence
The article does not mention any publicly announced licensing deals or revenue-sharing arrangements between Meta Superintelligence Labs and named creative entities or artists. While partnerships with cloud vendors (Microsoft, AWS, Google Cloud) and strategic alliances with NVIDIA and Scale AI are mentioned, no licensing or revenue-sharing arrangements specifically with creators are described.
Sources
- •Article Content - Corporate Structure and Products & Services sections
Provenance/attribution tooling for AI outputs
Provenance/attribution tooling for AI-generated outputs (C2PA, watermarking, metadata tagging).
Evidence
The article does not describe any provenance/attribution tooling, watermarking systems, C2PA metadata implementation, SynthID, or commitment to a specific provenance standard for AI-generated outputs from Meta Superintelligence Labs. While the article mentions generative AI tools for video and image generation, no disclosure of provenance or attribution mechanisms is provided.
Sources
- •Article Content - Products & Services section
Workforce impact assessment or commitment
Published workforce impact assessment or commitment (labor market effects, reskilling, human-in-the-loop programs).
Evidence
The article mentions that Meta has launched initiatives such as the 'Llama Impact Grants,' which provide funding and technical support to organizations using MSL technology to address social challenges in education, healthcare, and economic development. This constitutes a published statement naming a specific initiative. However, no detailed workforce impact assessment or quantifiable commitments regarding job displacement, skills retraining, or measurable outcomes from the program are provided.
Sources
- •Article Content - Societal Impact section
Does NOT claim ownership over user-generated outputs
ToS does NOT claim copyright/exclusive ownership over user-generated outputs. Silent ToS = 0.
Evidence
The article does not disclose Meta's Terms of Service regarding user-generated output ownership. While the article mentions that MSL provides tools for users to create and generate content through the Meta AI App and Studio, and that Meta asserts these tools empower the 'creator economy,' there is no explicit statement that users retain full ownership or have unrestricted rights to their generated outputs. The absence of disclosure regarding user output ownership rights results in a score of 0.
Sources
- •Article Content - Products & Services section
Governance
14/25Discloses corporate structure, investors, and board
Publicly discloses corporate structure, major investors, and board composition.
Evidence
The article discloses that Mark Zuckerberg serves as Chairman and CEO with direct oversight of the lab, Yann LeCun serves as Chief AI Scientist, Joelle Pineau is Vice President of AI Research, and Chris Cox is Chief Product Officer. The article also states that Meta Superintelligence Labs is a wholly-owned subsidiary of Meta Platforms, Inc., which is publicly traded on NASDAQ under ticker 'META,' and identifies that institutional investors including The Vanguard Group and BlackRock are the largest shareholders, collectively holding over 13% of shares. Both board/leadership and major investors are publicly disclosed through official channels.
Sources
- •Article Content - Corporate Structure section
Independent ethics/safety advisory board
Independent ethics/safety advisory board with verifiably external members. Internal trust & safety team alone = 0.
Evidence
The article mentions the Meta Oversight Board, described as an independent body that reviews high-stakes content moderation and policy decisions and has issued over 300 policy recommendations in its first five years. However, the article explicitly states that 'the board's effectiveness has been a subject of debate among independent observers, who have noted its tendency to defer controversial decisions back to Meta's executive leadership.' This indicates that while a body exists, its independence is unclear, and members' independence from Meta is questionable. The score reflects existence but unclear independence.
Sources
- •Article Content - Safety & Ethics section
Legal corporate structure preserving safety/mission
Corporate structure preserves safety/mission mandate via a legal mechanism (PBC, capped-profit, charter clause).
Evidence
The article does not disclose any special legal corporate structure designed to preserve safety or mission commitments. Meta Superintelligence Labs is described as 'a wholly-owned subsidiary of Meta Platforms, Inc., which is publicly traded on the NASDAQ,' suggesting a standard corporate structure. No mention of Benefit Corporation status, capped-profit structures, or other verifiable legal mechanisms designed to protect safety/mission is provided.
Sources
- •Article Content - Corporate Structure section
Public policy engagement or lobbying disclosure
Public policy engagement or lobbying disclosure: positions on AI regulation, lobbying spend, governance framework signatory.
Evidence
The article mentions regulatory attention from U.S. Senators Elizabeth Warren and Ron Wyden regarding Meta's 'acqui-hire' tactics, indicating Meta's awareness and engagement in policy discussions. However, no evidence is provided that Meta has published policy positions, signed onto specific frameworks, disclosed lobbying activities, or engaged in active public policy advocacy. The regulatory scrutiny directed at Meta does not constitute Meta's own public policy engagement or disclosure.
Sources
- •Article Content - Safety & Ethics section
No senior departures citing safety/ethics (last 36 months)
No publicly documented senior leadership (VP+) departures or whistleblower events citing safety/ethics concerns in the last 36 months. Only on-record statements count.
Evidence
The article does not document any senior (VP-level or above) departures from Meta Superintelligence Labs publicly citing safety or ethics concerns in the last 36 months. While the article mentions a 'significant brain drain' of researchers departing to competitors or independent ventures and cites a 'perceived loss of academic independence,' no specific named senior departures with on-record safety/ethics concerns are provided. The absence of documented departures on public record results in a clean record score.
Sources
- •Article Content - Reception & Controversies section
Sources
- 1Article Content - Safety & Ethics section
- 2Article Content - Research & Development section
- 3Article Content - Lead and Products & Services sections
- 4Article Content - Products & Services and Safety & Ethics sections
- 5Article Content - Reception & Controversies section
- 6Article Content - Reception & Controversies and Societal Impact sections
- 7Article Content - Corporate Structure and Products & Services sections
- 8Article Content - Products & Services section
- 9Article Content - Societal Impact section
- 10Article Content - Corporate Structure section
Scores are generated using the Amallo Ethics Rubric (Organisation v4) based on publicly verifiable information. Each criterion is scored against defined tiers — only exact tier values are valid. Evidence is sourced from official documentation, research papers, and independent analyses. Scores may change as new information becomes available.