Summary

• G7 leaders discussed closer cooperation on access to frontier AI after restrictions affected some foreign use of advanced Anthropic systems.
• Cybersecurity is a central part of the dispute because the same AI capabilities can support defenders and strengthen attackers.
• Model access is becoming a security, procurement, and resilience issue for governments, regulated businesses, and critical infrastructure providers.

The G7 is trying to preserve access to advanced AI among allied countries while acknowledging that the most capable systems now carry security risks normally associated with strategic technologies.

Leaders and technology executives have discussed a trusted partner approach to frontier AI access after restrictions on some foreign use of advanced Anthropic models raised concern among European governments. French President Emmanuel Macron has pushed for broader access, particularly where systems could support cyber defence and public interest use cases.

The dispute centres on AI systems capable enough to assist with cybersecurity work. Such tools can help defenders analyse code, test software, identify weaknesses, and respond more quickly to incidents. The same capabilities can also help attackers conduct reconnaissance, understand vulnerabilities, automate parts of an intrusion, or improve social engineering. That dual use character makes frontier AI difficult to treat as ordinary enterprise software.

The G7 agenda now places AI access alongside financial stability, productivity, labour markets, resilience, and cyber defence. AI has moved from a product category into the operating fabric of public administration, financial services, software engineering, infrastructure management, and national security.

European governments face an immediate access problem and a longer term dependency problem. Researchers, companies, and public institutions want to use the strongest systems for legitimate work. At the same time, public bodies and regulated sectors cannot build permanent processes around technologies that may become subject to sudden access restrictions, export controls, or national security decisions outside Europe.

Regulated businesses face similar constraints. A bank, insurer, telecoms operator, or energy company cannot assess frontier AI only on price and performance if the system is tied to cyber operations, compliance monitoring, software development, or customer workflows. Procurement teams need to examine jurisdiction, continuity of access, auditability, contractual rights, data handling, and exit routes.

A trusted partner framework could offer a compromise between open access and unilateral restriction. Approved countries, companies, and use cases might receive access to advanced systems under common safeguards. The model would resemble arrangements in export controlled technology and sensitive supply chains, although AI is harder to monitor because it is delivered through cloud platforms, updated continuously, and embedded into applications.

Those differences make enforcement difficult. A physical component crosses a border once. An AI system may be accessed thousands of times a day by users across departments, vendors, and customer facing services. Understanding who uses it, for what purpose, under which safeguards, and with what downstream effects requires much more than a licence agreement.

Competition will also shape the outcome. US model developers dominate much of the frontier AI market, while European businesses and public bodies are under pressure to adopt AI quickly. If leading systems become harder to access, European buyers may accelerate work with domestic providers, open models, or multi model procurement. That could create room for European AI companies, but only where their systems meet enterprise expectations for performance, security, support, and documentation.

The G7 talks show why AI governance cannot be separated from cybersecurity. Advanced systems can strengthen defenders, but attackers will use automation wherever it reduces cost and increases reach. Governments want to protect defensive access without giving hostile actors the same advantage, and that boundary will remain difficult to draw.

Trusted access will only be useful if organisations can plan around it. Businesses that treat frontier AI as a controlled strategic capability will design for model diversity, portability, stronger internal governance, and better visibility over how AI systems affect operational processes.