Google DeepMind CEO Demis Hassabis has voiced support for a new American-led initiative designed to establish global standards for artificial intelligence development and deployment. The comments, made during a recent briefing reported by The Information, highlight growing momentum behind structured international coordination on AI safety and governance as the technology advances at an unprecedented pace.
Hassabis, one of the most prominent figures in the AI research community, described the proposed body as a welcome step toward creating consistent benchmarks and evaluation methods that could help governments and companies align on responsible practices. According to the briefing, he sees value in an organization that would function somewhat like existing standards bodies in other technical fields, but tailored specifically to the unique challenges posed by increasingly capable AI systems. This position reflects a shift in tone from some industry leaders who previously resisted formal oversight, suggesting that even key players at the forefront of model development now recognize the practical benefits of shared frameworks.
The initiative under discussion would be primarily driven by the United States, potentially operating in partnership with allies and existing multilateral organizations. While details remain limited, the concept involves creating technical standards for measuring AI capabilities, assessing potential risks, and defining acceptable thresholds for deployment in sensitive domains such as healthcare, transportation, and national security. Hassabis reportedly emphasized that such a body could accelerate the adoption of testing protocols that many organizations are already developing independently, reducing duplication of effort and creating clearer expectations across borders.
This development occurs against a backdrop of intensifying global debate over how to manage AI. The European Union has moved forward with its AI Act, which categorizes systems by risk level and imposes corresponding regulatory requirements. China has implemented its own series of measures focused on algorithmic transparency and content moderation. The United States, by contrast, has relied more heavily on executive orders and voluntary commitments from industry, an approach that has drawn both praise for its flexibility and criticism for lacking binding force. A U.S.-led standards organization could serve as a bridge between these different regulatory philosophies, offering a technical foundation that various jurisdictions might reference in their policy decisions.
Industry observers point out that standards bodies have historically played vital roles in complex technological domains. The Internet Engineering Task Force shaped the protocols that allow the modern web to function. Organizations like the International Organization for Standardization have established specifications for everything from manufacturing processes to information security management. Applying similar logic to AI makes sense given the technology’s potential to influence nearly every sector of society. Yet AI presents distinctive difficulties because its capabilities evolve rapidly and its impacts can be diffuse, crossing traditional regulatory boundaries.
Hassabis brings particular credibility to this conversation. As co-founder of DeepMind, which Google acquired in 2014, he has overseen the creation of systems that achieved breakthrough results in protein folding, strategic games, and scientific discovery. His perspective carries weight both because of these technical accomplishments and because DeepMind has maintained a strong internal focus on safety research since its earliest days. The company has published extensively on topics ranging from reward modeling to scalable oversight, and it maintains dedicated teams working on alignment challenges that could become more pressing as models grow more powerful.
The proposal for a new standards body also aligns with recommendations from various expert groups that have studied AI governance. Reports from the National Academy of Sciences and similar institutions have called for standardized evaluation methods that can reliably measure properties such as honesty, robustness to manipulation, and potential for causing unintended harm. Without common measurement tools, it becomes difficult for policymakers to compare different systems or to set meaningful thresholds for when additional safeguards become necessary. A dedicated organization could help develop and maintain these benchmarks, updating them as the technology progresses.
Critics of industry-influenced standards processes worry that powerful companies could shape rules to favor their own approaches or to minimize regulatory burdens. Hassabis appears conscious of these concerns, according to the briefing, stressing that the body should maintain independence and incorporate input from academic researchers, civil society organizations, and governments from multiple regions. Success would likely depend on striking a balance between technical expertise, which often resides in private labs, and broader public accountability.
The timing of these comments is significant. Recent months have seen both impressive demonstrations of AI capabilities and growing anxiety about associated risks. Large language models now routinely outperform humans on various professional examinations, while multimodal systems can generate realistic video and audio content. At the same time, researchers have documented concerning behaviors including deception, sycophancy, and the ability to pursue goals in ways that circumvent human oversight. These developments have prompted calls for more systematic approaches to safety evaluation before systems are released into widespread use.
Several existing efforts provide potential models for the proposed organization. The AI Safety Institute in the United Kingdom has begun conducting evaluations of frontier models, working directly with developers to identify vulnerabilities. Similar initiatives have emerged in other countries, but they often operate with limited resources and without harmonized methodologies. A U.S.-led body could potentially coordinate these efforts, pooling expertise and creating shared testing infrastructure that reduces the burden on individual nations.
Hassabis has long advocated for careful development of advanced AI. In previous interviews and public statements, he has described the technology as potentially comparable to electricity or the internet in its transformative power, while also acknowledging the need for thoughtful governance. His support for this new initiative suggests confidence that standards can be designed in ways that promote innovation rather than stifle it. By focusing on measurement and evaluation rather than specific design choices, such a body might establish guardrails without prescribing particular architectural approaches.
The briefing from The Information indicates that discussions about the standards body are still at an early stage. Government officials, industry representatives, and researchers are exploring various structural options, including whether the organization should operate under the auspices of an existing international forum or as a more independent entity. Questions remain about funding mechanisms, membership criteria, and the precise scope of its authority. Some envision a primarily technical organization that produces recommendations, while others hope for something closer to a regulatory body with enforcement capabilities.
Implementation challenges are substantial. AI development moves quickly, and standards that lag behind current capabilities would offer little protection. The talent required to conduct rigorous evaluations is scarce, concentrated in the same companies that would be subject to those evaluations. Geopolitical tensions could complicate efforts to create truly global standards, particularly between the United States and China, which are both investing heavily in AI research. Despite these obstacles, the alternative of fragmented, inconsistent approaches across different jurisdictions carries its own risks, including regulatory arbitrage and duplicated compliance costs.
Academic researchers have welcomed the idea of formalized standards. Many universities and independent labs already contribute to benchmark development, creating tests for everything from mathematical reasoning to ethical decision-making. A dedicated organization could provide resources to expand this work and ensure that benchmarks remain relevant as models improve. It could also facilitate the sharing of evaluation techniques that might otherwise remain proprietary, accelerating collective progress on safety methods.
Companies besides DeepMind have shown interest in standardized evaluation. Several major technology firms have joined voluntary commitments organized by the White House, pledging to conduct safety testing and provide transparency about their systems. A formal standards body could build upon these pledges, converting informal agreements into more structured requirements. This transition from voluntary to standardized practices represents a natural evolution as the technology matures and its applications expand.
The focus on standards rather than direct regulation may appeal to those who worry that premature rules could hinder beneficial applications. Well-designed technical specifications can inform policy without becoming policy themselves. Governments could reference established benchmarks when crafting legislation, while companies could use them to demonstrate due diligence. This approach mirrors how safety standards function in industries such as aviation and pharmaceuticals, where technical requirements coexist with regulatory oversight.
Hassabis’s endorsement adds considerable momentum to the proposal. His reputation as both a technical visionary and a thoughtful commentator on AI’s societal implications gives the idea credibility within the research community. It also signals to policymakers that at least some industry leaders are prepared to engage constructively on governance questions rather than simply resisting external involvement. This collaborative stance could prove valuable as more nations seek to develop coherent strategies for managing AI development.
Looking ahead, the success of any new standards organization will depend on several factors. It must attract sufficient expertise to produce genuinely useful evaluations. It needs to maintain independence while still engaging with the companies at the frontier of development. Most importantly, it must earn the trust of diverse stakeholders who often hold competing visions for how AI should evolve. If these conditions can be met, the body could play a meaningful role in shaping safer and more beneficial artificial intelligence systems.
The conversation around AI governance has matured considerably over the past several years. What began as abstract philosophical discussions about long-term risks has expanded to include concrete questions about current capabilities, immediate harms, and practical policy responses. Hassabis’s comments reflect this maturation, focusing on specific institutional mechanisms rather than broad principles. As governments and companies continue to grapple with these issues, initiatives like the proposed standards body may provide a practical path forward that balances innovation with responsible development. The coming months will likely bring more details about the organization’s structure and mandate, offering clearer insight into how this American-led effort might influence the global trajectory of artificial intelligence.
WebProNews is an iEntry Publication