How Amazon’s CSO defends against efforts by North Korean IT workers to infiltrate his company
Steve Schmidt, the chief security officer at Amazon, says his team has identified and blocked more than 1,800 attempts by North Korea to secure IT roles at the tech giant. He warns that this scheme is becoming more prevalent across the technology industry as the nation-state actor targets the lucrative salaries of generative artificial intelligence and machine learning jobs, and the troves of valuable data such workers have access to.
“A lot of people don’t think about organized efforts by other parties to get people hired into organizations who have interesting data,” says Schmidt, speaking at an event held by Amazon this week. “It’s actually pretty prolific.”
Schmidt says that in 2025, Amazon has seen a 27% increase in the number of North Korean applications on a quarter-over-quarter basis.
Notable cases throughout the year that point to the growing issue include four North Korean nationals being charged for allegedly scheming to get hired as remote IT workers and then steal nearly $1 million in cryptocurrency; a campaign to create a fake job-application platform to get hired at major AI companies; and a woman in Arizona who was sentenced to eight years in prison for her role in a $17 million scam to help North Koreans steal U.S. identities to secure remote IT roles.
These identity theft schemes represent an ever-escalating confrontation between nation-state actors like North Korea and major Fortune 500 companies, as bad actors develop new deception techniques and businesses respond by bolstering their defenses. The cycle continues and escalates because, for countries like North Korea, these schemes can generate big financial windfalls and access to proprietary data.
AI is increasingly being used as a tool to monitor and identify these criminals, but also by the criminals themselves for attacks. Last month, Anthropic generated headlines when it disclosed that purported Chinese operators used that AI startup’s coding tool to target about 30 organizations.
Schmidt says the North Korean approach has changed over time, evolving from creating entirely fabricated profiles online to purchasing identities from Americans with legitimate backgrounds. The hackers will then aim to use these credentials to infiltrate an employer.
He says that Amazon has bolstered defenses through a mix of AI-enabled tools and human prevention efforts, a process he says the company has refined over the past two years. AI models have been trained to look for suspicious activity, including how North Korean operatives may list their contact information. They tend to use a plus symbol at the front of a phone number, which most Americans don’t do, and Amazon has identified around 200 different academic institutions that these IT workers use in their résumés.
These fake IT workers will also list nonexistent companies in their employment history. Some of these fake companies may actually have a registered business presence in a given state with a human who works for them to “verify” past employment, but they have no real operations.
Amazon now conducts more interviews in person and Schmidt says that the company’s mandate to bring workers fully back in the office also has some security benefits. “It is very, very hard to hide behind somebody else’s identity when you have to be in the office,” Schmidt tells Fortune.
Identity verification is now required at multiple stages throughout the interview process. And once someone is hired, Amazon keeps an eye on suspicious patterns of computer usage and the quality of work that’s being produced. Schmidt says the bad actors produce software code that is “markedly lower” in quality when working in the office versus when they are remote.
He calls for IT and human resources departments to more closely coordinate on hiring. At Amazon, the security team has access to the résumés, LinkedIn feeds, and other data that recruiters use to lure talent, and AI models are used to flag accounts that look suspicious. “It’s actually a lot cheaper for the HR organization if we discover the problem up front,” says Schmidt.
Amazon’s internally developed authentication system is called Midway; it both verifies an employee’s identity and controls access to their systems. The company relies on what’s known as “Universal 2nd Factor,” which uses physical security keys, rather than one-time passwords. Authentication requires a device that Amazon trusts, with the physical token and a pin that’s associated with that token.
Schmidt says Amazon’s security team is leveraging AI in quite a few ways, including speeding up security analysis (reviews that traditionally took hours and can now be completed in about 10 minutes); detecting and removing fake AI-written reviews on the company’s retail page; and identifying potential flaws in AI-written software code. The latter effort is called “autonomous threat analysis,” in which two sets of AI agents compete with each other to look for problems in the code and mitigate them before a product is launched.
As Amazon has embraced agentic AI capabilities, Schmidt says the company made an investment in Midway to build software that would allow it to securely identify the agent itself, as well as the action it has been authorized to take on behalf of a person. AI agents are like humans in that they need boundaries: An AI agent in robotics shouldn’t have access to the retail division, while a customer service agent shouldn’t touch Amazon Web Services.
“That agent that’s in the middle is not a service, which is the underlying layers of software talking to each other, and it’s not a human, it’s both together” says Schmidt. “We had to make that investment to ensure that we put the right boundaries around the agent.”
John Kell
Send thoughts or suggestions to CIO Intelligence here.
This story was originally featured on Fortune.com