EU Scientific and Ethics Advisors Urge Stronger AI Safeguards and Independent Model Tests

The EU's scientific and ethics advisors issued a rare joint statement calling for stronger AI safeguards and an independent EU capacity to evaluate models, rather than leaving that task to the companies that build them.

Christopher Ort
By Christopher Ort · 6 min read
Share
A quiet Brussels meeting room with stacks of unmarked documents on a conference table, pale daylight on glass and empty chairs, no flags or readable text.
Image: AI-generated, i10X
On this page
  1. Quick Take
  2. In plain English
  3. What happened
  4. Background
  5. Why it matters
  6. What's still open
  7. Sources
  8. i10x Perspective

The European Commission’s two independent advisory groups on science and on ethics have taken the unusual step of issuing a joint statement on artificial intelligence. In a Commission news release dated 16 November 2026, the Group of Chief Scientific Advisors and the European Group on Ethics in Science and New Technologies said the next phase of AI requires “heightened vigilance and effective safeguards,” and that “traditional guard rails no longer suffice.” They argue that Europe already has an ambitious legal framework, including the EU AI Act, but that those rules will fail if they are not enforced, kept strong, and matched by the EU’s own ability to evaluate models.

i10X Quick Take

Quick Take

  1. Who spoke: The Group of Chief Scientific Advisors, seven independent experts appointed in a personal capacity, and the European Group on Ethics, which advises the Commission on ethical, societal and fundamental-rights questions. A joint statement from both groups is unusual.
  2. The core warning: As AI becomes “increasingly powerful, autonomous and embedded in everyday life,” existing safeguards are not enough, and a lack of political commitment or institutional capacity could undermine obligations already on the books.
  3. Four recommendations: keep the rules strong, with safety, security and human agency as prerequisites; calibrate the pace of development and deployment to the degree of potential danger; build independent EU capacity to evaluate models, rather than leaving that task to commercial developers; and build an ecosystem for safety, security and continuous learning that keeps up with new capabilities.
  4. The legal backdrop: the AI Act (Regulation (EU) 2024/1689), the 2025 General-Purpose AI Code of Practice, and the Commission’s proposal to strengthen Europe’s cloud and AI ecosystem, the Cloud and AI Development Act, COM(2026) 502 final.
  5. What this is not: new legislation. It is advice to the Commission, not a vote, a regulation or a deadline.

In plain English

Europe already has a big AI law. Two official advisory groups, one made up of scientists and one focused on ethics, have now said that the law is not enough on its own. They want the European Union to keep the rules strong, to go slower on the most dangerous uses, and to build its own labs and methods for testing AI models, instead of relying on the companies that make those models to grade their own homework. They also warn that the rules could quietly weaken if politicians do not back them or if outside pressure succeeds in delaying them. This is a warning and a set of recommendations. It does not, by itself, change the law.

What happened

The Directorate-General for Research and Innovation published the news article on 16 November 2026, summarizing a joint statement titled “The next phase of artificial intelligence: Governing for safety, security and human agency.”

The advisors’ starting point is that AI “presents distinctive challenges for public policy.” They say Europe should not have to choose between innovation and safety, but that “growing capabilities” have to be “matched by stronger evidence, effective oversight and the capacity to identify, assess and manage risks.” In the Commission’s phrasing: “Regulation must evolve alongside the technology, not lag behind it.”

The enforcement warning is unusually direct for a scientific advisory note. The advisors “warn that a lack of political commitment or institutional capacity risks undermine existing obligations, while external efforts to delay or weaken legislation could further compromise safeguards.” The Commission text does not name the countries, companies or lobbying efforts it has in mind.

The most concrete institutional recommendation is about evaluation. “The EU must develop its own capacity to evaluate models, and avoid delegating this task to commercial developers.” That goes to the heart of how general-purpose AI rules are carried out in practice: who runs the tests, who has access to the models, and whose evidence a regulator is prepared to trust.

The other three recommendations are broader. “Keep the rules strong, with safety, security and human agency as essential prerequisites for AI development.” “Calibrate rates of development and deployment to degrees of potential danger.” And “build an ecosystem for safety, security, and continuous learning, with evidence and evaluation methods that evolve in step with AI capabilities.” Human agency and “meaningful human oversight” are treated as non-negotiable, not as optional add-ons.

Background

Both groups have advised the Commission on AI before, including on the use of AI in science, on AI in crisis management, and on governing “neuro-AI.” The Commission notes that a joint statement is an unusual step, which is why this one is being presented as a signal about the “next phase” of the technology rather than as a routine opinion.

The legal references in the release are worth unpacking. The AI Act is already law. The 2025 Code of Practice is the accompanying compliance tool for general-purpose AI. The Cloud and AI Development Act is still a proposal, COM(2026) 502 final, aimed at strengthening Europe’s own cloud and AI ecosystem. The advisors are therefore speaking into a mixed moment: one landmark law in force, a code that is meant to make it workable, and a new industrial proposal that is not yet adopted. Their message is that passing rules is not the same as having the people, tests and institutions to make those rules real.

This sits in a wider transatlantic argument about whether AI rules should be tightened, loosened or coordinated. The statement does not mention the United States, the United Kingdom or any company by name. It does, however, warn against “external efforts to delay or weaken legislation,” which will be read in Brussels as a reference to that argument, even though the text leaves the actors unspecified.

For a country-by-country view of how AI policy is taking shape, i10x maintains an AI country comparison that tracks national approaches alongside the EU framework.

Why it matters

Advice of this kind does not move markets, and it does not immediately change what a model provider must file or test. It does tell you what a large part of the Commission’s expert estate wants the next two years of implementation to look like: no weakening of the Act, more public evaluation capacity, and a tighter link between how dangerous a system could be and how fast it is allowed to spread.

The evaluation point is the one with practical consequences. If the EU builds a genuine independent testing capability, model providers selling into Europe will face an auditor that is not their own safety team and not a consultant they hired. If it does not, the advisors’ warning is that the Act’s most ambitious requirements will depend on evidence supplied by the same firms those requirements are meant to discipline.

There is also a political test. The advisors say a lack of “political commitment or institutional capacity” could hollow out rules that already exist. That is a polite way of saying the risk is not only that new laws fail to pass. It is that old ones are underfunded, understaffed or quietly relaxed.

What’s still open

  • Whether the Commission will answer the joint statement with a specific plan, budget and timetable for an independent EU model-evaluation capacity.
  • How the Cloud and AI Development Act proposal is amended, if at all, in light of the advisors’ emphasis on safety and oversight.
  • What “calibrate rates of development and deployment to degrees of potential danger” would mean in operational rules, and who would make that judgment.
  • Whether member states will fund the institutional capacity the advisors say is missing, or treat the statement as a one-day news item.

Sources

  • European Commission, Directorate-General for Research and Innovation, “AI requires rigorous safeguards and rules that keep pace with technological advances: experts tell the European Commission,” 16 November 2026
  • Joint statement by the Group of Chief Scientific Advisors and the European Group on Ethics in Science and New Technologies, “The next phase of artificial intelligence: Governing for safety, security and human agency”
  • Regulation (EU) 2024/1689 (EU Artificial Intelligence Act), the 2025 General-Purpose AI Code of Practice, and Commission proposal COM(2026) 502 final

i10x Perspective

The useful sentence in this statement is not the call for “heightened vigilance,” which every AI ethics board can write, but the refusal to outsource evaluation to the vendors. Europe has spent two years writing an AI rulebook. The advisors are now saying the hard part is the laboratory, the staff and the political will to use them. If the Commission cannot point to a funded evaluation capacity in the next budget cycle, this joint statement will look, in hindsight, like a well-argued memo that arrived after the enforcement gap had already opened.

Christopher Ort

Christopher Ort

Christopher Ort handles SEO, AI agent workflows and PR & partnerships at i10x.ai, the all-in-one AI workspace built by i10X Global Pte. Ltd. in Singapore. He is a graduate of Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU). He is the author of “The Style Penalty”, a study published on SSRN that analyses 1,576 evaluation decisions by four frontier AI models to show how AI-written CVs fare in automated hiring screens; the research has been covered by heise online and t3n.

Editor's picks

  1. Nokia, Nscale and Telia Form AI Gigafactory Northern in the Nordics and Baltics
  2. Gemini in Apps Script Is Generally Available for Google Workspace Automations
  3. JetBrains TeamCity Plugin Runs Codex and Claude Code Inside CI Builds
  4. Microsoft Research Open-Sources TokenEconomizer to Cut Coding Agent Costs by up to 31%
  5. Tencent Launches Hy4 Flash on TokenHub, Priced at $0.07 per Million Input Tokens