Tuesday · November 17, 2026

 ·  Daily  ·  Newsletters

Resonance network: Quantum InsiderSpace Insider

Policy & Government · AI Safety

US Tightens AI Chip Export Rules as FBI Warns of Autonomous AI Hacking Tools

An American flag with white stars on a blue field above red and white stripes

The US government is moving on several AI security fronts at once, with the Trump administration preparing new export controls targeting remote access to AI computing power and the FBI warning that hackers are using autonomous AI tools to carry out cyberattacks worldwide.

Under an interim rule set out on Monday, the Commerce Department will require licences to export, reexport or transfer AI chips and servers to data centres anywhere in the world if they may provide computing power to companies in China or other restricted countries. The rule targets companies based in the US’s D:5 group of embargoed countries, which includes China, Russia, Iran, North Korea and 18 others, closing a gap that allowed firms to bypass existing controls by accessing AI infrastructure located abroad.

The rule applies to data centres or networks supplying more than 12 million total processing performance (TPP), roughly the maximum computing power of about 750 Nvidia H100 or H200 chips, and covers indirect access through brokers or intermediaries. Exporters and data centre operators must monitor customers’ ownership structures and activities, and existing operators covered by the rule must obtain licences within 90 days. Commercial recommendation systems used in e-commerce and social media are exempt.

New licensing requirements will also apply to advanced chip shipments to Malaysia, Singapore, Indonesia, Thailand and Brazil, which the Commerce Department said carry a higher-than-normal risk of diversion, though chips destined for US-based companies are excluded. The rule formally rescinds the Biden administration’s January 2025 AI Diffusion Rule. The department described the new control as a logical extension of existing restrictions, intended to stop companies in countries of concern from using advanced US semiconductors to train and run models that could support military and intelligence activities. In line with recent federal guidance, the rule refers to AI as super intelligence, or SI.

The move follows the resignation of Nvidia co-founder Jensen Huang after he admitted to an affair with Alice Huang, a former executive at Nvidia customer Megaspeed whom US officials have alleged is a Chinese spy. The US government has reportedly sought to place Megaspeed on a special entity list over suspicions that it bought thousands of Nvidia chips to provide computing power to Chinese businesses. Reuters reported last week that Chinese AI firms were rushing to secure overseas capacity ahead of tighter US restrictions.

President Donald Trump told reporters on Monday that he does not believe US national security was put at risk by Huang’s affair. He questioned whether Alice Huang is a spy, saying it had not been proven, while adding that Huang probably should not have acted as he did on a personal level. Those remarks contrast with an internal memo reportedly produced under the oversight of Secretary of State Marco Rubio that warned she was working for China’s Ministry of State Security. Trump was reportedly briefed on those findings at least twice, including in May aboard Air Force One shortly before Huang joined him on a trip to China, though Trump has said he was unaware of the claims. He also said calls for Huang’s return were not a bad idea, after figures including Elon Musk, Garry Tan and Tim Sweeney signed a public letter supporting his reinstatement, and noted he had recommended Tim Cook for the role.

As well as this, the FBI has issued a flash advisory warning that hackers are using open source AI-powered penetration testing tools such as Artex to attack financial services, government, technology, defence, telecoms and legal organisations. Artex can be connected to frontier models from Anthropic and OpenAI as well as open source and non-US models, and can move from targeting a system to active compromise in under three hours. While its attacks generate large volumes of traffic that may trigger alerts, the FBI said the speed makes it difficult for human defenders to keep pace.

The tool typically scans networks for weaknesses, exploits exposed or stolen passwords and weak protections, and then spreads to databases and mailboxes. Artex has been used to breach South Korean financial institutions, exposing the personal data of at least 68,000 people, and to compromise IT management firm ConnectWise, giving attackers Microsoft says are likely Chinese access to the Microsoft 365 environments of at least 2,250 customers. Cybersecurity company Barracuda also disclosed a breach after the tool found an unrevoked vendor credential, prompting it to reset all customer credentials. Barracuda chief information security officer Arve Kjoelen said defenders who rely on spotting and responding to malicious activity will often arrive after attackers have finished.

Artex’s developer has since made the software closed source, but the FBI said earlier open source versions remain widely accessible, and researchers recently found hundreds of exposed Artex systems online without passwords.

James Dargan
About the author
James Dargan

James Dargan is a writer and researcher at The AI Insider. His focus is on the AI startup ecosystem and he writes articles on the space that have a tone accessible to the average reader.

The AI economy, every weekday morning

The daily briefing plus the weekly Scale-ups to watch edition. Free, no spam, unsubscribe any time.