Security Experts Warn Nearly Everyone in AI Is Open to Honey-Trap Schemes

Security specialists are warning that people across the AI industry are susceptible to honey-trap schemes, in which an intelligence operative builds a personal relationship with a target to extract information or access. TechCrunch reported the warning on November 20, 2026, drawing on comments from Aaron Brown, a former senior CIA officer and founder of software company Lumbra, and Perry Carpenter, a cybercrimologist and chief deception strategist at KnowBe4.
Brown said honey-trap attackers gain access by posing as investors, customers, recruiters and consultancies. Each of those roles is a normal part of how AI companies raise capital, win contracts and hire talent, which is what makes them useful as cover. Brown also told TechCrunch that nobody briefs founders on counterintelligence, and said the AI industry became a national security industry faster than its culture caught up.
Carpenter said AI can help sustain an entire fictional backstory for honey-trap schemes. The point is operational. A convincing persona once required sustained human effort to maintain across emails, calls and social media. Generative tools lower that cost, so an attacker can keep a cover identity consistent over a longer period than was previously practical.
The same report describes a specific set of accusations. Jensen Huang has been accused of having an affair with Alice Huang as part of an alleged honey-trap scheme. Alice Huang, a former executive of a Singapore neocloud (a cloud provider built around rented GPU capacity for AI workloads), has been accused of being a Chinese spy. She has also been accused of smuggling billions of dollars worth of restricted Nvidia chips into China. These are allegations, and the article reports them as such.
The Wall Street Journal reported that OpenAI started warning employees about outsiders trying to cozy up to them for company secrets. That report, as TechCrunch relays it, places the AI sector's concern about approaches of this kind inside a major lab's internal security practice, not only in the abstract.
In my view, the most useful part of this story for working engineers and founders is the mundane one. Honey-trap tradecraft does not require a breach of systems. It works through the ordinary channels of the business: a new investor who wants a deep technical briefing, a recruiter who asks about model training pipelines, a consultancy that wants access to a data room (a secure folder where confidential documents are shared during deals). Security controls built around networks and credentials do little against a relationship that looks like a routine business contact.
The broader context here is that the AI industry's growth has pulled it into the orbit of export controls, chip supply chains and state competition far faster than its hiring, board practices and internal training were designed for. Brown's remark about culture is a statement about that gap. Closing it is mostly a matter of practice rather than technology: explicit counterintelligence guidance for leadership, verification habits for unsolicited contacts, and clear reporting paths when something feels off. Those measures are unglamorous, and they are also inexpensive relative to what a single leaked roadmap or training run can cost.
The people who build and fund AI systems are now a target set for foreign intelligence tradecraft, and the defenses that protected earlier generations of tech companies were not built for that kind of approach. Companies that adapt their onboarding, investor diligence and staff briefings to that reality will be better placed to keep their work both productive and protected.


