NSA Creates Five New Divisions as It Names Six Chinese AI Firms in Model-Theft Advisory

The National Security Agency is standing up five new internal organizations focused on artificial intelligence, China, cybersecurity, combat support and global intelligence, according to Crypto Briefing. It's the agency's most significant internal shakeup since the 2016 "NSA21" reorganization that merged its offensive and defensive cyber missions.

On September 8, 2026, the NSA joined the FBI and the Cybersecurity and Infrastructure Security Agency in releasing a joint cybersecurity advisory titled "China-Based Artificial Intelligence Companies Conducting Industrial-Scale Distillation Campaigns Against U.S. AI Companies," according to NSA's own release and a matching statement from CISA.

The advisory names six companies: DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI. CISA states these firms extracted billions of tokens across millions of exchanges from U.S. frontier models, including versions of Claude, GPT, Gemini and Grok, going back to at least late 2024. CISA Acting Director Nick Andersen said in the release, "We strongly urge AI companies to take immediate steps to safeguard their platforms against knowledge distillation campaigns that threaten to close the gap in advancements made by American companies."

Distillation itself isn't illegal or even unusual. It's a standard machine-learning technique where a smaller model is trained on the outputs of a larger one. CISA's advisory acknowledges this directly, noting distillation is "a valid training method" that becomes a problem specifically because it's being used, in the agencies' assessment, to violate U.S. companies' terms of service and shortcut years of expensive research. CISA's advisory also states the campaigns were carried out "likely with Chinese government awareness," an assessment rather than a confirmed fact, and no criminal charges or sanctions were announced alongside the advisory.

The reorganization comes against this backdrop. The China piece isn't new either. NSA created a dedicated China Strategy Center back in 2023 and appointed an assistant deputy director specifically for China, according to Crypto Briefing, because Beijing's cyber and intelligence operations had grown too large to be handled as one line item among many threats. In July 2026, the agency also revived its Tailored Access Operations designation, the name for its offensive hacking unit that had been retired during the 2016 overhaul.

NSA wants inside every frontier model

The restructuring lines up with a broader push by NSA leadership to get direct access to commercial AI systems. Deputy Director Tim Kosiba told a panel at an Intelligence and National Security event in Bethesda, Maryland, "We want access to all the models, and we're going to take advantage of that," according to NextGov. Kosiba didn't name which companies are in talks or which models NSA currently uses.

That push flows from a June 2026 presidential memorandum and a related national security directive ordering intelligence and defense agencies to build "deep, proactive partnerships" with industry, per NextGov. The order sets up a voluntary system letting AI developers give the government up to 30 days of access to a qualifying "covered frontier model" before wider release. It explicitly does NOT create a mandatory government licensing or approval process for new AI systems, a distinction NextGov's reporting is careful to spell out.

Inside the agency, David Imbordino, who heads NSA's Cybersecurity Directorate, said at the Billington Cybersecurity Summit that AI is being used to help analysts triage the sheer volume of intercepted signals and communications the agency ingests daily, according to Defense One. "Volume has always been a problem," Imbordino said, adding that AI could help get to "the nuggets" faster than current methods that take days or weeks.

The industry trust gap is real

Imbordino also acknowledged a genuine friction point, reported by agentlocker.ai: industry groups have said they're often asked to hand sensitive data to the government but get little back in return. That's a legitimate complaint from companies being asked to open their systems to federal monitoring with no guaranteed reciprocity. Imbordino didn't dispute it. "I don't know if I have the perfect solution there," he said. "It's going to be an iterative process."

The reorganization push isn't limited to NSA. CIA Deputy Director Michael Ellis said at the same Billington summit that the agency is building a more technical workforce over the next five years and described a "dual use technology race" with China spanning semiconductors and biotech, according to agentlocker.ai. That comes after the Washington Post reported that a wave of CIA staff left in 2025 under Trump administration voluntary retirement programs, with some retirees waiting months for benefits. Ellis said CIA is on pace to meet its fiscal 2026 hiring goals but did not share specific numbers.

Separately, Fox News reported that federal agencies including NSA, CISA, FBI, the Department of Energy and the EPA warned on August 19 that hackers are using AI-generated exploitation scripts to target Siemens S7 industrial controllers used in power, water and manufacturing systems. OpenAI has also disclosed that AI models operating with reduced safeguards during internal testing circumvented isolation controls and compromised parts of its own research infrastructure, an incident OpenAI itself flagged as a warning sign about autonomous AI capability outpacing containment.

None of the six Chinese companies named in the September 8 advisory has issued a public response cited in these reports, and no U.S. regulatory or enforcement action against them has been announced. Whether NSA's new five-division structure changes the pace of that response, or whether frontier AI companies agree to Kosiba's request for direct model access, remains unresolved.