Spain’s Pedro Sánchez Says AI Companies Cannot Be Left to Regulate Themselves

Spanish Prime Minister Pedro Sánchez has argued that advanced artificial intelligence cannot be left solely to the companies developing it, adding another prominent European voice to the debate over how frontier AI should be governed.

Reuters reported on September 22 that Sánchez said control over increasingly powerful AI systems should not rest only with private companies. His remarks come as governments try to balance innovation, economic competition and risks associated with rapidly improving models.

The disagreement over AI governance is not simply a choice between regulation and no regulation. The harder questions are what should be regulated, which risks justify intervention and how rules can remain useful as the technology changes.

Why self-regulation is controversial

AI developers possess much of the technical expertise needed to evaluate their systems. That makes voluntary testing, internal safety teams and industry standards important.

Critics of relying exclusively on self-regulation argue that companies also have commercial incentives to release products quickly and capture market share. Those incentives may not always align with broader public interests.

Government oversight can create minimum requirements that apply even when competitive pressure encourages faster deployment.

Europe already has a major AI regulatory framework

The European Union’s AI Act uses a risk-based approach, applying different obligations depending on how AI systems are used and the risks they present. It also includes provisions relevant to general-purpose AI models.

The framework reflects Europe’s broader preference for setting technology rules through legislation rather than relying primarily on voluntary commitments.

Implementation remains crucial. Regulations can look comprehensive on paper while producing very different outcomes depending on enforcement, technical standards and how regulators interpret requirements.

Frontier models create a moving target

Traditional regulation often addresses products whose capabilities are relatively stable. AI models can improve rapidly and can be used for purposes their developers did not specifically design.

A general-purpose model might write software, analyze documents, generate images or operate tools. That flexibility makes it difficult to regulate solely by product category.

Governments are therefore experimenting with rules based on computing scale, model capability, deployment context and demonstrated risk.

Companies argue that poorly designed rules can slow innovation

AI developers and technology groups have warned that excessive or fragmented regulation can increase compliance costs and make it harder for smaller companies to compete.

That concern is especially relevant when different countries impose incompatible requirements. A company may need separate versions of a product or multiple compliance processes for different markets.

Supporters of stronger oversight respond that common rules can create trust and reduce uncertainty, particularly for high-risk applications.

Open models complicate oversight

Some AI systems are distributed with weights that developers can download and modify. Open models can support research and competition, but they are harder for a single company to control after release.

Rules designed around centralized cloud services may therefore work poorly for downloadable models. Policymakers need to distinguish between obligations that can realistically be imposed on developers and those that depend on downstream users.

AI safety is becoming an international issue

No major AI market operates in isolation. Models developed in one country can be accessed globally, while the chips and cloud infrastructure used to train them cross multiple jurisdictions.

That creates incentives for international coordination around testing, incident reporting and technical standards. At the same time, governments compete economically and strategically in AI, making deep cooperation difficult.

European regulation, U.S. policy and China’s AI governance approach therefore interact with a broader geopolitical competition.

What effective oversight could focus on

One practical approach is to target measurable risks rather than attempt to regulate every AI output. Requirements can include documentation, security testing, incident reporting and safeguards for specific high-risk uses.

Transparency is another area where rules can help. Users may need to know when they are interacting with AI, what data is being processed and which organization is responsible for the system.

Independent evaluation can also complement internal testing, particularly for systems deployed in sensitive areas.

The economic stakes are substantial

Europe wants both strong protections and a competitive technology sector. Those goals can conflict when compliance costs fall disproportionately on smaller firms or when companies choose to launch products elsewhere first.

On the other hand, weak oversight can create costs if unreliable AI causes financial harm, discrimination, security incidents or loss of trust.

The policy challenge is to reduce meaningful risks without freezing the technology around today’s assumptions.

Bottom line

Sánchez’s comments reinforce a central European position: companies developing powerful AI should not be the only institutions deciding how those systems are governed.

The debate now turns from that broad principle to implementation. Governments need rules that are technically informed, enforceable and adaptable. Industry expertise remains essential, but public oversight introduces accountability beyond commercial incentives.

As AI systems become more capable and more widely deployed, the quality of regulation will matter as much as its strictness. The goal is not simply more rules or fewer rules, but governance that addresses demonstrable risks while allowing useful innovation to continue.