‍

The June 2, 2026, Executive Order (EO) titled Promoting Advanced Artificial Intelligence Innovation and Security has charged a group of federal agencies with advancing US efforts on artificial intelligence (AI). Specifically, the Department of the Treasury, the National Security Agency (NSA), and the Cybersecurity and Infrastructure Agency (CISA) are to work with the Office of National Cyber Director (ONCD), the Assistant to the President for Science and Technology (APST), and the National Institute of Standards and Technology (NIST) in developing a process by which AI developers can engage with the federal government to select “trusted partners” to have early access to certain covered frontier AI models. (Sec. 3(b)(iii)). The purpose is to modernize government and private sector information systems and harden them against external threats; to protect innovation and intellectual property; and to cultivate advanced AI-enabled capabilities. The President’s AI EO adds rigor, transparency, and accountability to what is currently an ad hoc process.

‍

Establishing a test environment for technology is a well-worn path that allows third-party developers and other stakeholders an opportunity to test for vulnerabilities or unexpected issues that may arise before the technology is more widely available. For the results to be meaningful, though, this “sandbox” approach needs clearly defined parameters on which trusted partners can test the technology. The EO offers some guidance on this point, noting the AI industry and operators of critical infrastructure should coordinate and deconflict scanning for software vulnerabilities, discover and validate such vulnerabilities, and coordinate and prioritize remediation and distribution of vulnerability patches. What the EO does not offer guidance on, and what the respective federal entities will need to determine, is who in the critical infrastructure sectors should be included in frontier AI model early access.

‍

We believe four gating criteria should guide frontier AI model early access:

‍

1. Is the entity requesting access a critical infrastructure owner or operator? As the EO recognizes, critical infrastructure owners and operators are prime targets for cybersecurity and national security infiltrations. The communications sector, for example, has been designated a critical infrastructure sector because of its role in “enabling functions across all critical infrastructure sectors.” In other words, exploiting a vulnerability in a communications network could have negative cascading effects on other sectors. As such, communications providers, and in particular major broadband providers, should have the opportunity to participate in early access programs so that they can test and remediate vulnerabilities before bad actors can use the same frontier AI models to exploit these vulnerabilities.  

2. Does the requesting entity manage access to the full critical infrastructure stack? Not every communications provider needs (or wants) to participate in an early access program testing frontier AI models. The goal should be to focus access on those market participants with widest visibility into network vulnerabilities. Getting participants from all network layers, including core network (servicing colocation data centers), transport network, and the last-mile access network is important in evaluating the readiness of models for commercial deployment because each of these points in the networks have different vulnerabilities.  

3. Does the requesting entity have the necessary capabilities to fully assess and mitigate frontier AI model risks? Companies involved in this process should also be required to demonstrate a dedicated, internal workforce that has the skills necessary to inform testing and identify solutions based on their past experiences. AI frontier models are by definition operating at the cutting edge of AI advancement, but skilled network engineers, software engineers, cybersecurity experts, and others who understand how a company’s network operates can help quickly inform the assessment of these models and do so in a solutions-oriented way.  

4. Does failure to provide the requesting entity early access to the frontier AI model risk significant harm to the US economy? Early access to frontier AI models should focus on critical infrastructure entities that, if targeted by bad actors or are taken down due to technical issues, have the potential to cause substantial harm to wide cross-section of the US economy. The top five ISPs in the US – measured by number of subscribers served – cover between 70-75% of the fixed broadband market. The top three mobile operators in the US cover approximately 95% of the mobile connectivity market. Beyond last-mile network operators, ensuring coordination with colocation data centers and transport providers is important. In the US, the top three core network providers account for 60% (by revenue) of the market and similarly the top 5 Tier 1 transport network providers represent 75-80% of global internet routes. Which is all to say that while no telecommunications network operator or data center is too small a target for bad actors, it is clear that these larger network providers should receive the earliest access to frontier AI models, which will have downstream benefits for all market participants.

‍

Federal guidance is expected in the coming weeks and it is our hope that it embodies some of these principles. We know the stakes are high based on past experience addressing similar concerns in the early days of internet access deployment. It can feel like building a plane while flying it, but it is critical to act. The EO gives us a solid opportunity to do so in a way that helps protect not only our national and cyber-security, but our economic and personal security.