White House AI adviser attacks EU regulation at New Delhi summit

A senior White House official used a Tony Blair Institute event on the margins of the India AI Impact Summit in New Delhi to deliver a blunt assessment of European technology policy. Sriram Krishnan, appointed by President Donald Trump as senior policy adviser on artificial intelligence, told the audience that Brussels must abandon what he called "doomerism" and rewrite the regulatory atmosphere around innovation.

"I do think the atmosphere in the EU needs to change and be more focused on innovation, less focused on governance and less focused on doomerism," Krishnan said on Wednesday. His remarks, delivered before an audience of policymakers and industry figures, reiterated long-standing US opposition to the EU Artificial Intelligence Act, adopted in 2024 after years of negotiation between the European Parliament and the Council.

A regulation built on risk tiers

The AI Act classifies systems into four risk categories, unacceptable, high, limited and minimal, with obligations scaling accordingly. Providers of high-risk systems, covering areas from biometric identification to critical infrastructure, must meet requirements on data governance, transparency, human oversight and cybersecurity before placing products on the single market. The legislation also bans certain practices outright, including social scoring and real-time remote biometric identification in public spaces, with narrow exceptions.

For European officials, the framework represents a first-of-its-kind attempt to make AI trustworthy without banning the technology wholesale. Henna Virkkunen, the Commission's executive vice-president for tech sovereignty, security and democracy, has argued that legal certainty will ultimately attract investment by giving companies clear rules. But critics in Washington and parts of the European startup ecosystem contend the compliance burden falls disproportionately on smaller firms that lack the legal resources of Google, Microsoft or Meta.

The entrepreneur who left

Krishnan illustrated his argument with a single name: Peter Steinberger, the Austrian developer behind OpenClaw, a personal AI assistant platform. Steinberger, Krishnan said, is relocating to the United States to join OpenAI. The anecdote was offered as evidence that the EU's regulatory climate is pushing founders toward jurisdictions perceived as more permissive.

Whether one departure constitutes a trend is debatable. European venture capital data shows AI investment has continued to grow since the Act's political agreement in December 2023, though the gap with US funding volumes has widened. According to Dealroom figures, European AI startups raised roughly $12 billion in 2024, compared with more than $80 billion in the United States. Regulatory uncertainty is frequently cited by founders in surveys, but so are market size, talent depth and access to later-stage capital.

India positions itself as the alternative

Krishnan was notably warmer toward the host country. He praised India's regulatory approach as "pro-innovation", a phrase that has become shorthand in New Delhi for a strategy that favours voluntary guidelines and sectoral rules over a horizontal AI law. The Indian government has published a discussion paper on AI governance but has not tabled comprehensive legislation, preferring to work through existing IT rules and sector-specific regulators.

That positioning aligns with the broader trajectory of the AI summit series. Since the inaugural meeting at Bletchley Park in November 2023, where the UK convened governments and companies around a declaration on frontier model safety, each subsequent gathering has moved further from existential risk and closer to commercial deal-making. The Seoul summit in May 2024 produced a network of safety institutes; the New Delhi draft declaration, seen by Politico, does not contain the word "safety" at all.

A shifting diplomatic choreography

The absence of safety language is not accidental. French President Emmanuel Macron, who hosted the previous summit in Paris in February 2025, has been explicit about wanting France and Europe to capture more of the AI value chain. His government has backed domestic champions such as Mistral AI and pushed for EU industrial policy that complements the regulatory framework. Virkkunen's presence in New Delhi signals the Commission's determination to remain in the room where investment partnerships are negotiated, even as Washington signals impatience with the European rulebook.

The transatlantic tension is not new. During the AI Act's legislative phase, US officials, including then-commerce secretary Gina Raimondo, warned that overly prescriptive rules could fragment the global market and disadvantage American firms. The Biden administration nonetheless engaged constructively on the EU-US Trade and Technology Council, producing a joint roadmap on evaluation and measurement. The Trump administration has taken a markedly different tone, framing European regulation as a barrier to American competitiveness rather than a partner in standard-setting.

What the Act actually requires

Much of the criticism conflates the Act's provisions for general-purpose AI models, added late in the process at the Parliament's insistence, with its core high-risk regime. Providers of models deemed to pose systemic risk, a threshold tied to compute used in training, must notify the Commission, conduct adversarial testing, report serious incidents and ensure cybersecurity. For the vast majority of AI applications, however, the obligations are lighter: transparency for systems interacting with humans, labelling for deepfakes, and a code of practice for general-purpose models that is still being drafted.

The Commission has also established an AI Office to oversee enforcement, coordinate with national authorities and develop the codes of practice. Its staffing and budget remain modest compared with the US AI Safety Institute or the UK equivalent, raising questions about whether the regulatory state can keep pace with the technology it governs.

Talent, capital and the single market

The Steinberger anecdote touches on a genuine European vulnerability: the difficulty of retaining founders once they reach the scaling stage. The single market remains fragmented in practice, different labour laws, tax regimes, and venture capital ecosystems across 27 member states. The EU has responded with initiatives such as the European Innovation Council and the upcoming European Chips Act, but the gap in late-stage funding is structural. US pension funds and university endowments allocate far more to venture capital than their European counterparts, a disparity no regulation can fix.

Conversely, the AI Act's defenders argue that a unified European rulebook prevents a race to the bottom among member states and gives compliant companies a passport to 450 million consumers. The GDPR precedent is often cited: initially denounced as innovation-killing, it became a global standard that European firms now export as a competitive advantage.

The next test is implementation

The Act entered into force on 1 August 2024, with prohibitions on unacceptable-risk systems applying from February 2025. Obligations for high-risk systems phase in over 24 to 36 months depending on the category. General-purpose model providers face a 12-month deadline for the code of practice, which the AI Office is currently consulting on. How those codes are written, and whether they are proportionate for open-source developers and small labs, will determine whether Krishnan's prediction of regulatory strangulation bears out.