For the past few years, the global semiconductor industry has operated under the assumption that the United States was weaving an increasingly tight net around China. The narrative was clear: high-performance AI chips would be systematically blocked, and any entity aiding in their diversion would be swiftly neutralized via the trade blacklist. Developers and hardware architects watched as the US government tightened export controls, believing that the administrative machinery of the state was moving in lockstep with the urgency of the AI arms race. However, a critical fracture has appeared in this strategy, revealing that the net is not just fraying—it has effectively stopped growing.
The Administrative Freeze of the Entity List
The primary weapon in the US arsenal for controlling technology leakage is the Entity List. Managed by the Department of Commerce, this list serves as a high-stakes gatekeeper; any company placed on it is virtually barred from accessing US-made products, software, and core technologies unless they secure a specific government license. In practice, these licenses are rarely granted, turning a listing into a corporate death sentence for those reliant on American silicon. Yet, according to recent data, the addition of new entities to this list has completely stalled since October of last year. This represents the longest period of inactivity for the blacklist in over a decade, creating a vacuum in a period of unprecedented technological volatility.
This freeze is not due to a lack of candidates. The US government has already identified and vetted over 100 Chinese companies that meet the criteria for sanctions. These firms have already passed through the interagency committee process, receiving the necessary approvals for their inclusion on the list. Despite this, the Department of Commerce, which holds the final authority for official publication, has held the listings in limbo. Among the most prominent names waiting in this administrative purgatory are DeepSeek, a powerhouse AI research lab, and CXMT (ChangXin Memory Technologies), China's largest memory chip manufacturer.
The case of CXMT highlights the depth of this contradiction. The company has already been designated as a Chinese military company by the US Department of Defense. Furthermore, the Department of Commerce has been conducting a detailed review of CXMT for over a year to justify its placement on the Entity List. Every administrative box has been checked, and the security justifications are documented, yet the final step—the official listing—has not occurred. This delay is widely attributed to the current political climate of the Trump administration, which appears to be prioritizing the avoidance of immediate diplomatic escalation with Beijing over the rigid execution of security protocols.
The Security Cost of Diplomatic Hesitation
The gap between a security decision and its actual enforcement creates a window of opportunity that adversarial actors are already exploiting. While the US government hesitates to publish the names of these 100-plus firms, the companies themselves are not standing still. The delay has transformed a tool of national security into a loophole for technology acquisition. The most alarming example is DeepSeek. While the US government deliberates on its status, DeepSeek has been actively engaged in operations that undermine the very goals of the Entity List.
High-ranking officials from the US State Department have revealed that DeepSeek has provided critical support for Chinese military and intelligence operations. To sustain these efforts, the lab has not relied on legal channels but has instead constructed a sophisticated shadow infrastructure. By establishing paper companies across Southeast Asia, DeepSeek has created a buffer that allows it to illegally procure advanced US-made chips. These shell companies act as intermediaries, masking the final destination of the hardware and bypassing the export controls that the Entity List is designed to enforce.
Beyond hardware, the vulnerability extends to the AI models themselves. The industry has long viewed the use of an API as a safe way to provide services, but DeepSeek and two other Chinese AI labs have turned this into a vector for intellectual property theft. Anthropic recently confirmed a coordinated campaign where these entities attempted to illegally extract the core capabilities of the Claude AI platform. Similar targeting has been observed with OpenAI's models. This process, often involving model distillation or adversarial prompting to reverse-engineer logic, allows these labs to absorb the intelligence of US-developed frontier models and integrate that knowledge into their own domestic systems.
This creates a dangerous paradox. The interagency committee has already flagged at least 75 Chinese firms specializing in advanced semiconductor production and AI modeling for the blacklist, but Jeffrey Kessler, the Under Secretary for Industry and Security, has reportedly avoided the official listing. The hesitation stems from fears that formal sanctions will trigger a spike in US-China tensions toward the end of 2025. Consequently, the security apparatus is being overruled by trade policy concerns. As a result, the very companies the US has deemed a threat are currently operating with a level of freedom that the official policy suggests they should not have.
When the tools of regulation are paused for political leverage, the methods of evasion become more sophisticated. The current state of the Entity List proves that a list is only as effective as the speed of its execution. By allowing a backlog of 100 approved entities to remain unlisted, the US has provided a roadmap for Chinese firms to refine their third-country procurement routes and API extraction techniques. The effectiveness of a national security system is not measured by the number of names on a spreadsheet, but by the speed at which the door is closed once a threat is identified.
