Anthropic co-founder Jack Clark backs AI regulation standards

Debate over common rules for artificial intelligence intensified as systems moved from research settings into widely used products. Earlier technology governance often relied on a mix of voluntary industry practices, existing laws and sector-specific oversight. That approach became harder to assess as generative AI tools began producing text, images and code at scale, and as companies released more capable models to users across national borders. Governments and businesses faced a practical question: how could they judge whether systems were safe and trustworthy when companies used different testing methods and disclosed different information?

International efforts to establish shared principles predated the recent surge in generative AI. In 2019, the OECD adopted principles calling for AI that is innovative and trustworthy and respects human rights and democratic values. The G7’s Hiroshima AI Process in 2023 produced guiding principles and a voluntary code of conduct for organizations developing advanced AI systems. That same year, governments meeting at Bletchley Park issued a declaration recognizing the need for international cooperation on risks associated with advanced AI. These initiatives did not create a single global regulator or uniform legal obligations, but they helped establish a common vocabulary for discussing safety, transparency and accountability. At the same time, national and regional approaches began to diverge in their legal force and scope. The European Union’s AI Act, approved in 2024, established a risk-based framework with obligations that vary according to how AI systems are used and the level of risk involved. In the United States, federal policy during the same period emphasized measures such as safety testing, reporting and standards development, while leaving many questions to existing agencies and legislation. Other governments explored their own rules and guidance. This patchwork made international alignment consequential: companies operating in multiple markets could face overlapping requirements, while users and public institutions might have difficulty comparing safeguards. Anthropic was founded in 2021 by former OpenAI employees and became one of the companies developing large-scale generative AI models. Jack Clark, a co-founder who leads the company’s policy work, has participated in public discussions about how governance should develop alongside technical progress. The debate is not simply whether to regulate AI, but which responsibilities should fall on developers, deployers and governments, and how rules can be made meaningful without preventing beneficial uses. Measures such as evaluations before release, monitoring after deployment, disclosure of capabilities and incident reporting are among the approaches discussed by policymakers and researchers; their effectiveness depends in part on consistent definitions and credible implementation. Clark’s argument for clear standards therefore sits within a broader shift from general principles toward questions of implementation. Shared expectations can help governments compare risks and can give businesses and the public more confidence about how systems are assessed. Yet standards alone do not settle disagreements about enforcement, the pace of regulation or the balance between innovation and precaution. The continuing challenge for international AI governance is to make rules understandable and comparable across borders while adapting them as the technology and its uses change.