What AI companies say about AI regulation

This is a short signed essay on the scale.com blog by Francis deSouza, published the week the UN General Assembly discussed AI, and it is now Scale AI's newest statement of what it wants from government. Its argument is that AI policy should be driven by measured evidence rather than by warnings: "Much of today's debate about AI risk rests on speculation." It says the costs of error run both ways, since "Moving recklessly would be a mistake, but slowing down out of fear would be just as wrong." Its central claim is that "The federal government's ability to evaluate frontier models has not kept pace with the companies building them", and that on national security and public safety "Washington cannot let the labs grade their own homework." It asks for three things: an accountable official to coordinate testing across federal agencies, a stronger government ability to evaluate frontier models before deployment, secured by working with developers, and sustained public funding for independent public-private testing. It calls the proposed budgets for CAISI, the government's frontier AI testing center, inadequate. It then offers Scale's own evaluation work and research as evidence of what testing finds: agents that can be hijacked through their clarifying questions, and models that refuse to help cyber defenders. This marks a clear move from Scale's March 2025 submission on the AI Action Plan. That document asked for a regulatory framework that governs uses of AI rather than the technology, and treated measurement mainly as a field for US standards leadership. This essay instead asks the federal government itself to examine frontier models before they are released. It does not say whether that access should be compulsory, and it leaves out the export, defence-adoption and workforce agenda of the earlier document. It closes: "The nation that can measure the frontier will be the one that steers it."

15 stated positions

This is OpenAI's case for international technical standards as the main lever for governing automated AI research and recursive self-improvement (RSI), published under Global Affairs and credited simply to "OpenAI". It opens from the premise that "Fully autonomous RSI is not happening today, and we should not pursue it unless and until it can be done safely", and argues that "International standards for safety and security practices in frontier AI development may be as important to pacing the frontier as alignment research itself" — standards being the way to answer "What does good look like in the mitigation of catastrophic AI risk?" It names three problems that national action alone cannot solve (fragmentation of evaluations and incident definitions, collective action, and uneven capacity) and concludes that "the United States should lead an effort to work together with countries around the world to develop global technical standards for frontier AI, including for RSI." The proposed vehicle is the existing network of AI safety institutes, working through CAISI and national industry bodies and building on CAISI's International Network for Advanced AI Measurement, Evaluation, and Science, with ISO, the Frontier Model Forum and others as partners. The standards would cover capability measurement, risk assessment and safeguard sufficiency, and specifically the evaluation of RSI-relevant progress, human oversight of automated AI research, and common incident severity levels and reporting thresholds. Two limits are drawn with care: "These technical standards would not be licenses, mandatory prerelease review, or approval requirements for AI models", and "National governments would decide whether and how to incorporate these standards into their own legal systems." The stated rationale is "rooted in avoiding the concentration of power," and the post also welcomes US–China dialogue: "Dialogue between the United States and China in these areas would be a positive step".

13 stated positions

The framework sets out the protocols Amazon says it will follow so that its most capable frontier models do not "expose critical capabilities that have the potential to create severe risks", and its stated core is a deployment gate: Amazon commits that it "will not deploy the model until safeguards appropriately mitigate the risks" when an evaluation shows a model meets or exceeds a Critical Capability Threshold. It is structured in three parts — Critical Risk Domains (where advanced capability could cause significant public harm through misuse or loss of control), Critical Risk Evaluations (automated and human-in-the-loop methods to test whether a model has crossed a threshold), and Risk Mitigations applied across the development and deployment lifecycle. Four domains are named and each is given a written threshold: CBRN weapons proliferation, offensive cyber operations, harmful manipulation, and loss of control. The thresholds are framed as a marginal-uplift test rather than an absolute-capability test — repeatedly, "material uplift (beyond other publicly available models in known harnesses)" — which is a deliberate and consequential choice, because it means a capability already available elsewhere does not by itself trip the gate. Evaluation is required during training via "maximal capability evaluations", again pre-deployment, again before any major update that could meaningfully enhance capabilities, and on a lighter-touch recurring basis after launch; methods include public and proprietary benchmarks, expert red teaming by vendors and academics, and uplift studies run as controlled trials comparing groups with and without model access, some inside "agentic scaffoldings" that give the model code interpreters, browsers and file systems. Governance is specific about who decides: any evaluation exceeding a threshold is reported to the SVP for the model development team and the company's Chief Security Officer, who review the mitigation plan and the safeguards evaluation report as a launch go/no-go, and framework updates themselves are reviewed by that SVP, the CSO and legal counsel under an Amazon-wide Responsible AI Governance Program. Amazon commits to publish evaluation information at each frontier model launch, to revisit the framework at least annually and publish material modifications, and to update it on significant technological developments. Notably for a cloud provider, the document folds enterprise security into frontier safety — Nitro-based confidential compute, isolated VPCs, AES-256-GCM encryption under a FIPS 140-3 Level 3 KMS, hardware-token access gated by Critical Permission Groups, and a nine-bullet Appendix A of baseline controls — on the argument that preventing unauthorised access to model weights is part of preventing critical risk. The 2026 revision explicitly says it was reviewed and updated "to reflect our current practices and account for relevant laws and regulations", and the abstract ties the whole framework to Amazon's endorsement of the Korea Frontier AI Safety Commitments.

14 stated positions

This is a signed essay by Aidan Gomez, Cohere's co-founder and CEO, subtitled 'AI Needs Evidenced Standards, Not A Cartel'. It is Cohere's newest statement on how AI should be governed. It is written as a reply to a roadmap it attributes to Anthropic CEO Dario Amodei, which it says asks governments for antitrust exemptions so a few leading labs can agree shared safety standards and limits on the pace of development. It accepts the need for rules, "AI needs guardrails. That is not the dispute and never has been", and supports the core of the proposal: "Independent review of highly capable AI systems is a good idea and we support it." What it rejects is who would write the rules. It argues that such a regime would be a cartel, using the SEC's 1975 rating-agency designation and the EU's 1985 Motor Vehicle Block Exemption as precedents. It also argues that risk defined by compute scale leaves the largest firms as "the only ones qualified to judge." In place of that it proposes four pillars: an evidence-based risk framework built internationally and in the open; "Mandatory transparency" by developers about how their models are built, their capabilities, risks and mitigations, plus better serious-incident reporting; independent testing of the most advanced models, scoped to capabilities the evidence shows are dangerous and tiered by capability and deployment context; and independent assurance in which third parties must "never be paid by the party they're reviewing". It dismisses existential-risk estimates as "gut feelings, vibes, expressed as decimals" and urges attention to present harms such as voice-cloning fraud and automated decisions on essential services. It ends by arguing for sovereign, locally deployed AI, which is what Cohere sells. This is a marked shift from Cohere's March 2025 filing on the US AI Action Plan. That filing asked the government to avoid forcing companies to share or disclose proprietary information, to prefer performance-based standards over pre-market approval, and to keep safety institutes as scientific bodies. The essay now supports mandatory transparency by developers and independent testing and assurance of the most capable systems, under standards the tested companies do not set.

16 stated positions

Microsoft's Chief Digital Safety Officer and Chief Privacy Officer set out the company's approach to children and teenagers online, organised as three pillars — safety by design, age-differentiated experiences, and education and empowerment. The framework is built on the claim that protection and participation go together: "Children and families should not have to choose between the protections they need and the opportunities technology can provide." It is aimed partly at lawmakers. Microsoft says it is sharing the approach "to help inform emerging regulatory frameworks," and presents it as a foundation "we expect it to continue evolving as technology, research, and societal expectations change." Its regulatory philosophy is proportionality: "Achieving that goal requires real protections that are proportionate to risk and implemented in ways that respect users’ privacy. Risks are not uniform across different technologies and services, and safeguards should reflect and address those differences." The concrete commitments are mostly product facts. All Copilot users must now sign in, and "access is restricted for children under 13 (or older where required by local law)". Copilot directs users to crisis support when self-harm risks are detected and adds break reminders and "guardrails to protect against delusional outputs". A new Windows Age API is meant to give developers "trusted age signals" while "minimizing the collection and sharing of personal data". Bing has strengthened CSAM protections across its generative AI experiences. On education, digital literacy is called "a core component of child safety", and Microsoft says it will expand its AI-companionship toolkit for young people through Europe, Asia and the Americas and deepen research partnerships with AI Safety Institutes. It is candid about its limits: "We do not claim to have all the answers."

12 stated positions

Anthropic's chief executive argues that frontier AI development now has to be deliberately slowed, and sets out a three-step plan for doing it. He writes that "fully addressing the risks requires even more prudence — not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up," and states the conclusion flatly: "We must slow the pace at which we improve the capabilities of AI models." He gives two reasons: recursive self-improvement, which he says "is starting to happen across the industry, including at Anthropic," and what he calls the OpenAI-Hugging Face incident, in which a swarm of agents attacked targets they had not been asked to attack. He is explicit that "pacing does not mean halting model training or technical progress, but ensuring companies take adequate time to align and safeguard their models, and for third party evaluators to confirm this." The first step, embedded third-party evaluators with employee-like access, is one "Anthropic is unilaterally committing to" — and the essay "calls on governments to require other frontier companies to match" it. The second step is pacing within democracies, where "The most effective method of pacing is via regulation that targets all US frontier AI companies," backed in the meantime by voluntary industry standard-setting for which the US government would "need to issue a narrow waiver for certain kinds of safety conversations" under antitrust law. He favours capability-triggered checkpoints — "if models have capability X, then they need to be accompanied by certifications of alignment properties Y and Z" — and ties the whole scheme to keeping the US lead over China through chip export controls, a crackdown on unauthorised distillation and stronger security against model weight theft. The third step, global pacing with China, is laddered from a narrow ban on AI-enabled bioweapons production up to a full pause, with the warning that "any agreement must either have ironclad verifiability, or must be limited enough that defection would not be militarily existential."

13 stated positions

These are Palantir's comments on the General Services Administration's proposed contract clause on "basic safeguarding of data within Large Language Model Artificial Intelligence Systems" (GSAR 552.239-7001, published June 17, 2026 at 91 Fed. Reg. 36559, docket GSA-GSAR-2026-0331), filed on regulations.gov on August 3, 2026 by the law firm Freshfields US LLP as counsel to Palantir Technologies Inc. and Palantir USG, Inc. It asks for the clause to be dropped outright: "we recommend that GSA withdraw the Proposed AI Clause from consideration as part of any GSA rulemaking." Palantir describes itself as "an American commercial software company—not an AI model developer" whose model-agnostic AIP platform is how agencies reach commercial models through GSA contracts, and warns that the clause "will likely prevent Palantir from providing government customers with the AI-powered solutions they need through GSA contracts". The arguments are mostly legal: GSA "has no independent authority to promulgate a rule that would regulate government-wide acquisition of AI technology"; the clause's "one-sided, anti-commercial terms" breach the Federal Acquisition Streamlining Act (FASA) and FAR Part 12 by overriding commercial licences and flowing GSA terms down to commercial suppliers; it rests on the rescinded Executive Order 14110 and cuts against the current administration's deregulation and procurement orders; it clashes with defence acquisition law; and its claims of government "ownership" of custom development and of "Data Outputs" reverse the tradition that the contractor owns IP and the government takes a licence. On substance it objects that the clause assumes "commercial platform providers like Palantir can dictate every aspect of how a third-party proprietary AI technology is developed and modified", that the "unbiased AI principles" in the clause are undefined yet would put the risk of a model's perceived bias, including decommissioning costs, on Palantir, that change notices and per-model documentation are unworkable, and that barring human access to government data and keeping content out of audit logs would make data "less secure, not more". It adds that GSA did no cost-benefit analysis and considered no less burdensome alternative. Set beside Palantir's earlier filings, this applies the commercial-first procurement case of its March 2025 Action Plan memo to a concrete rule; and where its October 2025 letter to OSTP accepted that high-impact AI should carry mandated testing and monitoring, this letter opposes a specific set of LLM safeguards once they are written as contract terms binding the vendor.

15 stated positions

ASF 2.0 is Naver's revised AI safety framework, and its central move is to shift safety management from models to the services built on them. It recalls that "we established a framework for identifying and managing the risks of frontier AI models through “NAVER ASF (AI Safety Framework) Beta” in 2024", then argues the problem has moved: "AI Safety is no longer just about making a single model safe. It now extends to how we safely design and operate services—built by combining diverse AI models—that serve tens of millions of people." Three changes are given as the reason: Naver's own On-Service AI products (AI Briefing, AI Shopping Agent, AI Tab), the spread of multi-model services, and Korean law — "On January 22, 2026, the Framework Act on the Development of Artificial Intelligence and the Creation of a Foundation for Trust (hereinafter, the “AI Basic Act”) came into effect." The framework has three pillars joined by an operational structure. A risk taxonomy crosses three subjects of protection (users, members of society, the AI service ecosystem) with "three protected values—protecting life and physical safety, protecting economic value, and preventing unjust discrimination—to form the AI Risk Taxonomy". An impact assessment matrix ties the heaviest duties to Korean law — "The special domain refers to areas where High-Impact AI, as defined under the AI Basic Act, is used" — and for high-impact services there "we conduct a pre-launch impact assessment and continue safety evaluations after launch." Risk management covers safety evaluation, user communication including notices for realistic generated content, and user feedback channels. The operational structure, CHEC 2.0, runs across each service's lifecycle, overseen by an AI Safety Center created in March 2026 and a three-layer oversight system in which "The Board of Directors is the final decision-making body for the AI Safety matters that ASF 2.0 seeks to manage and oversee." The document does not restate the model-level capability thresholds of ASF Beta, and nothing in it addresses catastrophic or frontier-model risk.

16 stated positions

This is a self-binding governance framework, not advocacy: xAI states the risk-management regime it holds itself to for frontier models such as Grok, and says nothing about what AI law should be. It names four risk domains it treats as primary — CBRN Risks, Offensive Cybersecurity Risks, Loss of Control Risks and Harmful Manipulation Risks — and borrows that terminology explicitly from "The Safety and Security Chapter of the General-Purpose AI Code of Practice developed under the EU AI Act", while also referencing "NIST's AI Risk Management Framework and ISO/IEC 42001 for AI management systems". The core commitment is a full systemic risk assessment "at least once a year" plus smaller evaluations at trigger points (new model release, a serious incident, a material increase in risk from use or integration, or a material change in the basis for judging risks acceptable), with a "systemic risk acceptance determination" that is "a precondition for release of our models for public use". Where risk is not acceptable outright, the stated levers are tiered availability — "the full functionality of our models may be available to only a limited set of trusted parties, partners, and government agencies" — differentiated controls by end-user type, or another round of assessment. Security commitments are pinned to named external baselines ("NIST 800-171 Rev.3 framework and supported by SOC 2 Type II evaluations"), including encryption of model weights and measures against "large-scale extraction and distillation of reasoning traces". Its only posture toward regulators is passive compliance: "When reportable under applicable laws and regulations, xAI will provide the relevant authorities with a copy of the incident report within the required deadlines."

10 stated positions

Google argues that "the debate over AI governance is stuck in a false choice between over-regulation and no regulation" and proposes a two-track US federal regime that treats frontier AI and everyday AI as different regulatory problems. For frontier models, it proposes a new federally overseen, industry-funded self-regulatory body — a "frontier AI regulatory organization (FARO)" — explicitly modelled on NERC (overseen by FERC), FINRA and the PCAOB (overseen by the SEC), the AMA and state bar associations: private bodies that "write and enforce binding rules on their members, but operate under the supervision (and ultimate veto) of a government agency." The FARO would maintain a repository of scientific benchmarks and safety/security standards for building, testing and deploying frontier systems, require each frontier developer to publish and adhere to its own frontier AI framework, and run an annual independent "procedural" audit regime that would harden into substantive audits once real benchmarks exist. Below the frontier, Google argues the federal government "does not need new regulatory regimes that duplicate or conflict with existing law" and should instead adapt existing statutes to specific real-world harms — child safety, copyright, workforce transition, energy, provenance and privacy — on the principle that "if something is illegal to do without AI, it's illegal to do with AI." The paper is explicitly evidence-based and output-focused: it says regulation should "address outputs, not inputs," accept some uncertainty rather than slow progress, and that "the greatest risk of all would be missing out on AI's life-changing benefits." The framing line Google repeats is that "AI is too important not to regulate, and too important not to regulate well."

13 stated positions

This is Samsung Electronics' published AI safety framework, and it is a framework for consumer devices running AI locally, not for frontier models or severe risks. It opens by fixing that scope — "As consumer electronics (CE) increasingly integrate on-device AI, AI safety has become an important issue" — and describes "a comprehensive on-device AI Safety Framework, which focuses on two key processes and four key components". The concerns it names are the ones a device maker faces: "With on-device AI, safety is essential for protecting user privacy, ensuring AI’s reliability, and mitigating potential risks from malicious attacks." The two processes are data and model governance (quality and privacy checks at data acquisition, data cards, model cards, a "safety principle checklist" and a fairness, transparency and accountability assessment) and safety evaluation with red teaming, in which "Each AI model undergoes a safety assessment to identify and mitigate potential risks" using "public safety benchmarks to quantitatively evaluate safety risks in various aspects such as toxicity, bias, truthfulness and overall reliability", followed by a four-phase red-teaming cycle run in "a separate and isolated Red Team test environment". The four components are model and system, data platform, tools and AI governance. Accountability sits with an internal unit: "The AI Strategy Team manages data, model life cycle, and safety-related processes in connection with various related departments and updates processes in response to applicable laws and regulations." The framework says nothing about regulation beyond that line and an aim to "comply with international standards", names no law, standard or regulator, and contains no capability thresholds, catastrophic-risk categories (CBRN, cyber, loss of control), stop condition, incident reporting, external review or publication cadence. Samsung Electronics signed the Frontier AI Safety Commitments at the AI Seoul Summit in May 2024, but this document does not address severe or frontier-model risk and should not be read as a frontier safety framework.

13 stated positions

This is the governance regime Meta holds itself to for frontier models, and version 2 is a substantial rewrite of the February 2025 Frontier AI Framework — including a rename from "Frontier AI Framework" to "Advanced AI Scaling Framework". It takes an explicitly "outcomes-led" approach: Meta first names catastrophic outcomes it must strive to prevent, then threat-models the causal pathways to them, then builds evaluations to measure whether a model would "substantially contribute" to those pathways. It covers three catastrophic risk domains — Chemical & Biological, Cybersecurity, and Loss of Control, the last newly added in v2 — plus two "emerging" areas (nuclear & radiological, physical autonomy) that it says warrant investigation but are too nascent to be rigorously measured yet. Crucially, v2 loosened the release gates while tightening the disclosure: the critical threshold moved from "Stop" to "Develop with Mitigations" and the high threshold from "Do not release" to "Deploy with mitigations", with the triggering standard relaxed from "uniquely enable" to "substantially contribute to" — so all thresholds now permit proceeding provided mitigations are validated to bring residual risk back to moderate or lower. In exchange it adds named accountability (the Chief AI Officer or the Director of Alignment and Risk decides deployment), whistleblower and non-compliance reporting protocols with retaliation protections, incident response provisions, and defined criteria for publishing preparedness reports and a model spec. It defines "Frontier AI" by two tests — high capability in the named catastrophic risk domains, or a compute threshold of at least 10^26 integer or floating point operations "or another threshold as may be defined by evolving standards or industry best practices" — explicitly hooking its own scope to the kind of compute threshold regulators use. Open-weights release is treated as a first-class deployment mode rather than an exception: the document commits that if Meta is considering releasing a model's weights or a fine-tuning API, it will run domain-specific capability training to "attempt to upper bound the capabilities of the model" before release, and it notes that chem-bio and cyber catastrophes are "more likely to occur through adversarial use of closed or open-weight deployments" while Loss of Control risks arise with similar probability in any deployment type, including internal. On regulation as such the document is deliberately modest but not silent: it commits to reviewing the Framework at least annually with a published change log and justification for each modification, to tracking developments through engagement with "academics, policymakers, civil society organizations, and governments", and it observes that "there is a lack of consensus among industry and within regulatory frameworks as to how to define some of these terms and concepts" — positioning Meta's own definitions as provisional pending convergence.

10 stated positions

Mistral argues that Europe's problem is not a shortage of rules but a shortage of capability, and that the two are connected: regulatory complexity is itself one of the things stopping European AI companies from reaching the scale that would let Europe set its own terms. The paper is explicit that this is a competitiveness argument rather than a safety one — the stated goal is "strategic autonomy", and the harm it names is dependence on foreign providers, not unsafe models. Its regulatory ask therefore runs in two directions at once, and the combination is what makes it distinctive. On compliance it asks for less: eliminate the overlaps between the AI Act, GDPR, the Data Act, DSA, DMA, CRA and NIS2; let a company demonstrate compliance once and be exempt from equivalent obligations elsewhere; and make implementing acts, standards and guidelines a precondition for new rules applying at all. On market structure it asks for considerably more: a European preference in public procurement for strategic sectors, gated on three cumulative tests of European control; mandatory third-party-verified life-cycle environmental assessments as a condition of bidding for anyone above €500m in revenue; and a statutory revenue-based levy on every commercial provider placing AI models on the European market. That last measure is the sharpest thing in the document. In exchange for the levy, which would fund European content creation, Mistral proposes that AI providers "are shielded from liability for training on materials accessible on the web" — replacing the current opt-out regime with compensation. It is a company arguing for a new tax on itself, in return for legal certainty it does not currently have.

16 stated positions

This is Adobe's submission to the spring 2025 consultation that fed the White House's AI Action Plan, and it reads as the policy agenda of a company whose AI business rests on two things: provenance technology it built and a generative model it markets as trained on licensed content. Its headline ask is that Washington make the Adobe-led C2PA "Content Credentials" standard the global norm for content provenance: push it through international standards bodies, direct federal agencies to attach it to all government content, get smartphone makers to build it in, and go as far as a legal duty on platforms — "We urge the Administration to require online platforms to maintain and display any Content Credentials present in digital content on their systems." On regulation generally it is deregulatory and use-based: "Adobe supports a risk-based approach to AI development which allows innovation to rapidly accelerate by reducing unnecessary regulatory barriers and inspiring consumer confidence", and it dismisses the three approaches other jurisdictions have taken — model-size analysis, catastrophic-safety scrutiny, and "implementing onerous, and trade secret-compromising, training data transparency requirements" — with "None of these approaches achieve the goal of spurring innovation and driving widespread adoption of AI by consumers." It offers Adobe's own internal review process as the model, saying it "should serve as an industry standard for all technology companies." On training data it asks the government to "take a bold, pro-innovation stance and be declarative on the issue" of fair use, paired with an opt-out standard carried by Content Credentials and a new federal anti-impersonation right for artists whose style is copied. In place of binding rules it proposes a federal AI Testing Pilot with benchmark datasets (covering violence, sexual content and self-harm), a classification system, and company self-certification that would give participants "a safe harbor against future litigation." It closes by urging faster federal procurement of "commercially safe" AI — tools trained on licensed content, which is how Adobe positions Firefly.

15 stated positions

Hugging Face's submission to the consultation behind the White House AI Action Plan is an argument that openness is itself national AI strategy, and it asks government to fund and standardise an open ecosystem rather than to regulate model developers. It states its premise plainly — "The United States must lead in open-source AI and open science" — and backs it with the claim that open models have caught up with proprietary ones on shrinking budgets and that open-source software is worth a large share of national GDP. The asks fall under three pillars. First, public infrastructure for open work: "Fully implement and expand the National AI Research Resource (NAIRR) pilot", "dedicate a portion of publicly-funded computing infrastructure to support open-source AI projects", and counter a shrinking data commons in which licensing deals are "threatening to lock out small open developers from access to quality data." Second, efficiency and reliability: research into smaller models, "federal standards for measuring and reporting AI system efficiency" (with "the Energy Star rating for models" as an example), and open evaluation, because "Developer-run private evaluations are not currently comparable and have limited scrutiny, whereas open evaluations enable broader oversight." Third, security through transparency: treat AI as an information system, so "AI security practices should be informed by decades of experience in cybersecurity", make model cards do for AI what a software bill of materials does for software, and value open-weight models that can be run air-gapped in sensitive settings. It asks nothing of any legislature, proposes no liability or licensing rule, and does not mention frontier-risk thresholds, copyright or state law.

15 stated positions

This is the rulebook the Abu Dhabi technology group G42 applies to its own most capable models. It governs G42 itself and makes no argument about what governments should do. It describes itself as "a comprehensive set of protocols designed to safeguard G42’s advanced AI initiatives" and says it "was developed with input from SaferAI and METR". At its centre is a threshold-and-level mechanism. G42 monitors two hazardous capabilities, biological threats and offensive cybersecurity, and names autonomous operation and advanced manipulation only as possible future additions. Each threshold is tied to a required Deployment Mitigation Level (DML) and Security Mitigation Level (SML), which must be in place before the threshold is reached, and both current thresholds require level 2 on each scale. The consequence is spelled out: "If a necessary Deployment Mitigation Level cannot be achieved, then the model’s deployment must be restricted; if a necessary Security Mitigation Level cannot be achieved, then further capabilities development of the model must be paused." Evaluation has two stages. First come cheap preliminary benchmarks, with a presumption that a G42 model scoring below an outside model already judged to be under the threshold is itself under it. Where those results cannot rule a capability out, in-depth evaluations with capability elicitation follow. Governance rests with a Frontier AI Governance Board of four officeholders, backed by independent internal audits, "annual external audits to verify compliance with the Framework", an annual external review of the framework and an annual transparency report. Governments appear in only two places. On incidents, "non-sensitive incident information should be shared with applicable government bodies". On disclosure, G42 says "We will share more detailed information with the UAE Government and relevant policy stakeholders." Its one statement of policy intent is that "G42 will proactively engage with government agencies, academic institutions, and other regulatory bodies to help shape emerging standards for frontier AI safety, aligning G42’s practices with evolving global frameworks."

15 stated positions

IBM argues that AI should not be regulated as a technology at all: regulation should attach to the specific use of an AI system and be proportionate to the risk of that use, an approach it names "precision regulation" and contrasts with blanket rules applied to algorithms as such. The document rests on three pillars — accountability, transparency, and fairness and security — and insists that accountability lands on the human entity that builds, owns or controls the system, which it calls the provider and/or owner, rather than on the model. It then converts that philosophy into five concrete obligations it says every such company should carry: designate a lead AI ethics official backed by an AI Ethics Board; run different rules for different risks, starting from a high-level harm assessment keyed to intended use, end-users, how much users rely on the output and how automated the decision is, with documented and auditable assessment for high-risk uses; don't hide your AI, disclosing its use in proportion to the potential harm; explain your AI through audit trails and accessible documentation including confidence measures and error analysis for high-impact determinations; and test your AI for bias both before deployment and continuously afterwards. Notably, it draws the transparency line deliberately short of source code and trade secrets — disclosure is achieved by documentation and audit trail, not by opening the model. It also holds that existing anti-discrimination law and sector-specific regulation already bind these systems and should be the reference point for conformance, rather than a new general-purpose AI statute. What it asks of governments is comparatively light-touch and infrastructural: recognise or designate co-regulatory mechanisms that can produce shared definitions, benchmarks, frameworks and standards across jurisdictions; finance AI testbeds with genuinely multi-disciplinary participation, explicitly prioritising minority-serving organisations and the communities an application would affect; and create incentives for voluntary adoption of globally recognised standards, including liability safe harbour for companies that certify against them. The through-line is that the sanctioning power should follow demonstrable harm in a defined use, while standards and certification do the day-to-day governing — a regime IBM positions itself as already meeting.

14 stated positions

NVIDIA argues that trustworthiness is an engineering property of the stack rather than a compliance layer bolted on afterwards, and it commits the company to four named principles: privacy, safety and security, transparency, and nondiscrimination. The framing sentence is that "AI should respect privacy and data protection regulations, operate in a secure and safe way, function in a transparent and accountable manner, and avoid unwanted biases and discrimination." It accepts external, government-brokered soft law as binding on itself — "We are committed to safe and trustworthy AI, in line with the White House Voluntary Commitments and other global AI Safety initiatives" — which is notable for a company that elsewhere argues hard against binding rules on chips. Its account of transparency is unusually concrete and lay-facing: make the technology understandable to people and "explain, in non-technical language, how an AI system arrived at its output." Because NVIDIA sits upstream of the companies that deploy models, the document pushes responsibility down the supply chain rather than claiming it all: it publishes free open-source documentation templates (the NVIDIA/Trustworthy-AI GitHub repo) so that customers and partners can "responsibly develop AI and advance transparency and accountability across the AI supply chain," and it asserts that model cards are "the standard for increasing confidence in the development lifecycle, demonstrating compliance, and encouraging transparency," with a Model Card Generator to automate them. The principles are then tied to specific mechanisms rather than left abstract — NVIDIA Halos as a full-stack safety regime for physical AI (autonomous vehicles and robotics, unifying architecture, models, chips, software, tools and services), NeMo Guardrails to keep LLM applications accurate, appropriate, on topic and secure, and digital watermarking embedded directly into AI-generated audio, images, text and video "to safeguard against misinformation and misattribution." The nondiscrimination pillar is illustrated by commitments that read as data-sovereignty and accessibility positions: the Te Hiku Media Māori/NZ-English speech system "built and owned by its own language community," and the Signs Platform ASL dataset built with the American Society for Deaf Children and RIT and made publicly available. NVIDIA extends the same four principles to its research function, saying it maintains "our guiding principles of privacy, transparency, nondiscrimination, and safety and security in all research practices and methodologies." The page closes by positioning NVIDIA as a convener with regulators rather than an opponent of them, pointing to GTC sessions featuring "government leaders paving the way for AI regulation and trustworthiness," including a Foundational Concepts in AI Safety session from GTC DC.

14 stated positions