Bridging the Gap: The EU Wants to Regulate AI but Needs OpenAI and Anthropic to Let Regulators Through the Door

The European Union has positioned itself as a global leader in technology regulation. From the General Data Protection Regulation (GDPR) to the Digital Services Act, Brussels has shown it is willing to lay down rules for the digital world. Now, the EU wants to regulate artificial intelligence. But there is a catch: to write effective rules, regulators need to understand the systems they are regulating. And that is where the trouble begins.

As reported by the-decoder.com on May 11, 2026, the EU is running into a hard wall when it comes to accessing the inner workings of the most advanced AI models. The companies at the center of this tension are OpenAI and Anthropic — two of the most powerful AI labs in the world. Their models, including GPT variants and Claude, are the state-of-the-art in generative AI. But they are also largely closed. And that creates a fundamental problem for any regulator who wants to peek under the hood.

This article explores the heart of this issue. We will look at what the EU wants, why it needs access, how OpenAI and Anthropic are responding, and what this means for the future of AI for businesses, governments, and everyday people.

The Core Problem: Regulation Requires Access

To regulate something, you first need to understand it. That sounds obvious, but in the world of modern AI, it is anything but simple. The EU's upcoming AI Act, which is among the first comprehensive legal frameworks for artificial intelligence, aims to categorize AI systems by risk level. High-risk systems will face the most scrutiny. But how do you determine risk if you can't test the model?

The EU wants regulators to be able to examine AI models for bias, safety, robustness, and potential for misuse. This means they need access to the model itself, or at least to detailed information about its training data, architecture, and decision-making processes. Without this, any regulation is essentially based on trust rather than evidence.

OpenAI and Anthropic are not small startups. They are the leading forces in generative AI, and their models power everything from chatbots in customer service to code generation tools for developers. They have good reasons for keeping their models closed: protecting intellectual property, preventing misuse, and maintaining a competitive edge. But from a regulator's point of view, a closed model is a black box. And you can't set rules for a black box without opening it first.

How Did We Get Here? A Quick Look at the AI Landscape

To understand why this tension is so strong, we need to step back and look at the AI landscape as it stands in mid-2026. Over the last few years, generative AI has exploded into the mainstream. Tools like ChatGPT, Claude, and others are used by millions of people every day. They write emails, generate images, summarize documents, and even help with medical diagnosis.

This rapid adoption has caught the attention of governments worldwide. The EU, true to its tradition of digital rights protection, has been crafting the AI Act for several years. The law aims to ensure that AI is "human-centric" and trustworthy. But the law is only as good as its enforcement. And enforcement requires oversight.

OpenAI and Anthropic have both made public statements about the importance of AI safety. OpenAI has its own internal safety team and has published research on alignment. Anthropic was founded with a mission focused on safe and beneficial AI. But both companies have also been criticized for their lack of transparency. Their most powerful models are not open-source. They are available through APIs, but the inner workings remain proprietary.

What the EU Actually Need from AI Companies

The EU's request is not unreasonable. Regulators want to be able to perform audits. Here are some of the key things they would need to check:

Training data transparency: What data was the model trained on? Was it obtained lawfully? Does it contain biases that could lead to unfair outcomes?
Model behavior testing: Can the model be tricked into generating harmful content? Does it respect safety boundaries?
Decision-making explainability: Why did the model give a particular answer? Can its reasoning be traced?
Robustness against attacks: How does the model handle adversarial inputs designed to manipulate it?
Impact on fundamental rights: Does the model infringe on privacy, non-discrimination, or other rights?

To answer these questions, regulators need more than just documentation. They need technical access. In some cases, they may need to run their own tests directly on the model. This is where the friction with OpenAI and Anthropic becomes most acute. These companies fear that granting such access could compromise their trade secrets or create security risks. But from the EU's perspective, without access, the law is toothless.

OpenAI and Anthropic's Stance: A Pragmatic but Defensive Position

Neither OpenAI nor Anthropic has outright refused to cooperate with regulators. Both have engaged in dialogue with the EU and other governments. However, they are pushing for a model of regulation that relies more on self-reporting and external audits by approved third parties, rather than direct government access to the models.

The companies argue that they have their own safety measures in place. OpenAI has a "Preparedness Framework" and uses techniques like reinforcement learning from human feedback (RLHF) to align models with human values. Anthropic has published research on "Constitutional AI," a method for training models to follow a set of principles without requiring massive amounts of human feedback.

But regulators are skeptical. They point out that these internal safety measures are opaque and unproven from an independent perspective. A 2025 incident where an AI model was jailbroken to provide instructions for creating harmful substances showed that even the best safeguards can fail. Without independent testing, how can the public trust that such failures are being caught and fixed?

The tension is not unique to the EU. The United States, the United Kingdom, and other countries are also grappling with the same problem. However, the EU's approach is the most advanced in terms of legislation, making it the battleground for this fight.

Broader Implications for the Future of AI

This standoff has major implications for everyone involved — not just regulators and AI labs, but also businesses that use AI and the general public that interacts with it.

The Future of Oversight Might Be Open Models

One possible outcome is that the difficulty of regulating closed models pushes governments to favor open-source AI. If a model's training data, architecture, and weights are publicly available, regulators can inspect them at will. This would be a huge advantage for open models like those from Meta (Llama), Mistral, and others. We could see a shift in the AI market toward transparency, with "regulated AI" becoming a selling point for businesses that want to avoid legal trouble.

API-Level Regulation May Become the Norm

An alternative path is that regulators focus their attention on the interface — the API — rather than the model itself. Instead of demanding access to the underlying model, they could set rules for how the API is used, what data is logged, and what kind of auditing is performed by the company. This would be less invasive for AI labs but would still provide a degree of oversight. It is similar to how regulators oversee financial services: they examine the outcomes and processes without necessarily requiring access to every algorithm.

Compliance as a Competitive Advantage

For businesses that rely on AI, this situation creates both risk and opportunity. Companies that use models from OpenAI or Anthropic may face additional compliance burdens if their AI usage falls under high-risk categories. They may need to demonstrate that their AI provider meets EU standards. This could lead to a two-tier market: vendors that offer "regulatory-compliant AI" and those that do not. The big question is whether OpenAI and Anthropic will invest in becoming compliant, or whether they will cede that market to more transparent competitors.

Ripple Effects on International Policy

The EU's approach is likely to set a global benchmark. Just as GDPR became a de facto standard for privacy, the AI Act could shape how other countries approach AI regulation. If the EU succeeds in forcing transparency from AI labs, other countries may follow suit. If it fails, it could embolden companies to keep their models closed and argue for self-regulation. This battle is therefore not just about the EU; it is about the future of global AI governance.

Practical Insights for Businesses

So, what should businesses do now? The situation is still fluid, but there are some actionable steps to consider.

Audit your AI supply chain: Know which AI models your company uses and whether they can be audited by regulators. If you rely on closed models, start preparing for the possibility that you may need to switch to more transparent options.
Invest in AI governance: Even if regulators don't force you to, good AI governance is good business. Document how you use AI, what data it sees, and how you ensure it is safe and unbiased.
Watch the open-source trend: Keep an eye on models like Llama and Mistral. As regulation tightens, these may become the preferred choice for businesses that want to stay ahead of compliance requirements.
Engage with policy: If your business relies heavily on AI, consider contributing to public consultations on AI regulation. The EU has a history of engaging with stakeholders, and your voice can help shape a more practical framework.

The Societal Dimension: Why This Matters to Everyone

Beyond business implications, this is fundamentally about trust. AI is already making decisions that affect people's lives — who gets a loan, who gets hired, what medical advice they receive. If those decisions are made by black boxes that no regulator can inspect, we are putting a lot of faith in a few companies. History has shown that such concentration of power, without oversight, can lead to problems.

The EU's demand for access is not about stifling innovation. It is about ensuring that innovation happens within a framework that protects fundamental rights. The ball is now in the court of OpenAI and Anthropic. They have the opportunity to lead by example and propose a workable model for transparency that satisfies both safety and commercial interests. If they refuse, they force a confrontation that could ultimately limit their own market access.

Conclusion: The Door Must Open

The headline says it all: The EU wants to regulate AI but needs OpenAI and Anthropic to let regulators through the door. Right now, that door is only slightly ajar. The future of AI regulation hangs in the balance. Will we see a world where the most powerful AI systems are open to external scrutiny, or will we continue to rely on the promises of a few companies?

The next year will be critical. The EU AI Act is moving toward implementation, and the pressure on AI labs will only increase. The smartest move for OpenAI and Anthropic may be to open up, not because they are forced to, but because it is the only way to build the public trust that will sustain their businesses in the long run. A model that is trusted by people and governments is a model that will be used everywhere. A black box, no matter how powerful, will always face suspicion.

In the end, the door needs to open from both sides. Regulators must craft rules that are sensible and not overly burdensome. AI companies must show that they are willing to be held accountable. If they can find that middle ground, we could enter a golden age of safe, transparent, and beneficial AI. If not, we risk a future of fragmented regulation, stalled innovation, and a crisis of trust.

TLDR: The EU wants to enforce its AI Act but cannot properly audit the most advanced AI models from OpenAI and Anthropic because they are closed and proprietary. Regulators need technical access to check for bias, safety, and compliance. This creates a high-stakes standoff that will determine whether future AI governance is based on transparency or self-regulation. Businesses should prepare for a shift toward more open, auditable AI systems.