On November 9, 2026, Bruegel recommended that the European Commission bring together a coalition to create a shared international minimum for reporting and exchanging information about AI failures. The recommendation outlines a possible future arrangement; it does not establish an operating pact. Bruegel suggests a coalition pilot in early 2027. Read Bruegel’s proposal.

The idea addresses a practical problem: an AI system used in several countries could fail in one place before users elsewhere learn about the same risk. A shared reporting framework could help spread that information across borders. It would be separate from the reporting duties already set out in the EU AI Act.

What the proposed system would share

Bruegel’s proposed “common floor” would pair national reporting infrastructure with shared protocols and formats. Participating countries would exchange incident information on an ongoing basis, with verification and incentives intended to support participation. Under the suggested incentive model, countries that report less actively would receive less information from the shared pool.

The proposal does not call for countries to adopt another nation’s norms. Bruegel also says the design should guard against security risks. Those choices matter because a system meant to circulate incident details would need to encourage reporting while managing the risks of sharing sensitive information.

Bruegel illustrates the potential value with a hypothetical: a mortgage-scoring application used by banks in Berlin, Rio de Janeiro, Singapore and Washington produces biased decisions. If information about the failure reached users in other locations, they could be alerted to a similar problem. The scenario illustrates the intended cross-border function; it is not a reported incident.

For a cross-border AI failure, a shared reporting floor could help users in other countries learn about the problem sooner.

How the EU AI Act’s reporting duties differ

The EU AI Act already sets out serious-incident reporting duties, but they cover defined groups and direct reports to designated authorities. They do not create the proposed international information pool.

Under Article 73 of the AI Act, providers of high-risk AI systems report serious incidents to the market-surveillance authority in the EU member state where the incident occurred. The general deadline is immediately after the provider establishes a causal link, or a reasonable likelihood of one, and no later than 15 days after the provider becomes aware of the incident. The article sets a two-day maximum for specified widespread infringements or serious incidents, and a ten-day maximum for death-related cases, subject to the conditions in each provision.

Article 55 covers a different group: providers of general-purpose AI models with systemic risk. They must track and document relevant serious incidents and report the information, along with possible corrective measures, to the AI Office and, where appropriate, national competent authorities without undue delay.

The distinction is practical: Article 73 addresses serious incidents involving high-risk AI systems and sets circumstance-dependent deadlines; Article 55 concerns relevant incidents involving systemic-risk general-purpose AI models and requires reporting without undue delay. Neither provision establishes a shared international pool.

The proposed coalition’s path and limits

Bruegel names India, Canada, the United Kingdom and Brazil as possible coalition partners. It suggests taking the initiative to the United Nations and G20 after a possible pilot, then offering the reporting format to China and the United States. These are proposed steps, not commitments by the countries or forums named.

Bruegel presents incident-sharing as an initial step. It says the mechanism alone would not stop competition over frontier AI models or mitigate existential risks.

Scientists call for international oversight of advanced AI