Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
Applying export controls to AI models calls for public rulemaking, not undisclosed directives.
American export controls have long governed hardware, software, and controlled technology. The core categories covered by export controls, however, still assume that an identifiable item or controlled piece of knowledge is being shipped, transmitted, or released. In June, the U.S. Department of Commerce applied that framework to a service that users access remotely. Through a nonpublic directive, the Department reportedly required the artificial intelligence (AI) developer Anthropic to obtain a license before any foreign national could access its two most capable AI models. Reuters described the action as the first of its kind, and the episode marked a novel assertion of export-control authority over access to a commercially deployed frontier AI model.
The consequences arrived within hours. Because the order took effect immediately and Anthropic said that it had no reliable way to verify nationality in real time, the company could not immediately provide access only to eligible U.S. users while excluding foreign nationals. It suspended both models for all users, cutting off American users and allied institutions along with those covered by the directive. On June 30, the controls were lifted, and access returned the next day. The reversal resolved the immediate disruption but left the underlying legal questions unresolved.
Start with the most basic question: What, in this setting, is an export? Under the Export Administration Regulations, an export includes an actual shipment or transmission out of the United States. The “deemed-export rule” extends that logic to include a release or transfer of controlled “technology” or source code to a foreign person in the United States, treating that transfer as an export to the recipient’s most recent country of citizenship or permanent residency. Remote access to an AI model hosted on a cloud server does not fit either category cleanly. Users query an AI model running on the provider’s servers, but they do not receive the model weights or source code. The Center for Strategic & International Studies has argued that it is unclear when access to a remote model amounts to the release of software or technology under the Export Administration Regulations.
The procedural vehicle deepens the problem. According to public analyses of the nonpublic directive, the Commerce Department reportedly used an “is-informed” letter. Bureau of Industry and Security guidance explains that such letters can impose supplemental license requirements on named recipients without publishing a rule. The Commerce Department’s letter to Anthropic was not made public, and Anthropic said that it did not provide specific details of the national-security concern. The path to restoration was worked out through private negotiations, safeguard changes, and government testing. The Center for Strategic & International Studies noted that the episode leaves industry without a general framework for anticipating when similar authority will be used again. A regulated party that cannot know the rule in advance cannot design for compliance, and a market that cannot know the rule cannot price the risk.
Administrative law has a familiar mechanism for generally applicable standards: Notice-and-comment rulemaking exists so that binding obligations of general consequence are developed in public view, subject to scrutiny by the affected parties and the public and articulated in a form that courts can review. An is-informed letter can be lawful and useful for exigent, party-specific concerns, but the problem is one of fit. The June action was party-specific in form, but its practical effect was a worldwide change in the availability of a general-purpose commercial service, reaching users who were not the object of the specific security concern that prompted the measure. Anthropic said that its technology did not support a reliable, real-time nationality filter, so the company could not implement a narrower step immediately. When a recurring standard can predictably produce consequences far beyond the named recipient, the case for developing that standard through a published process becomes stronger.
The fix does not require abandoning the security interest. To the extent existing statutory authority permits, the Commerce Department could use notice-and-comment rulemaking to define when remote access to a hosted model constitutes an export or other controlled activity, state the capability thresholds and risk findings that would support a license requirement, and establish a review pathway with defined timelines. If existing law does not reach the remote-access problem cleanly, the U.S. Congress can clarify the authority rather than leaving the boundary to company-specific letters. Publishing the framework would not require disclosure of sensitive threat assessments; it would tell developers what facts matter, allied governments what to expect, and courts what standard they are reviewing. The June episode itself points toward a workable evidence process that can converge on testable safeguards through private negotiation and be translated into a durable public framework, with the added benefits of predictability and legitimacy.
Frontier AI models will keep improving, and the government will face this decision again, likely soon and with higher stakes. The question is not whether Washington may act against genuine security risks in commercial AI. It is whether the next action will rest on a rule whose scope is publicly knowable and a process affected parties can navigate or on another letter that no one outside the dispute can read.
Burak Oktenli is an independent researcher in AI governance and safety-critical autonomous systems.
States have the means to stop data center developers from skirting EPA energy regulations.
Scholars assess the risks and benefits of artificial intelligence in education.
AI labs should proactively adopt the transparency rules that a future U.S.–China agreement would require.
