Microsoft says North Korean agents use AI to deceive Western firms into hiring them

Microsoft’s threat‑intelligence blog details how Pyongyang has weaponised generative AI to streamline a long‑running recruitment fraud. The unit, which labels the actors “Jasper Sleet” and “Coral Sleet,” says the groups begin by prompting large‑language models to produce culturally appropriate name lists—e.g., “create a list of 100 Greek names”—and matching email‑address conventions. Those outputs seed synthetic identities that are then bolstered with AI‑generated headshots created via Face‑Swap, allowing the scammers to submit polished CVs that pass superficial checks. According to The Guardian, the operatives also scrape job boards such as Upwork for software‑development listings, using the posted skill requirements to tailor applications that appear technically credible.

During remote interviews, the agents employ voice‑changing software to mask their Korean accents, a tactic Microsoft says lets them sound like native English speakers. The blog post notes that once hired, the fake workers route their wages back to the North Korean state and, if terminated, have threatened to leak sensitive corporate data. To maintain the façade, they rely on AI to draft emails, translate documents, and even generate code snippets, thereby masking performance gaps that might otherwise expose the deception.

Microsoft quantifies the scale of the operation: last year the company disrupted roughly 3,000 Outlook or Hotmail accounts linked to these counterfeit IT workers. The threat‑intel team also observed that the scammers use AI to scan job postings for keyword matches, then automatically populate application forms with the generated names, emails, and images. Upwork, a major platform cited in the report, has responded by “taking aggressive action to remove bad actors,” though the blog warns that the AI‑driven pipeline makes detection increasingly difficult.

The advisory concludes with practical mitigations. Microsoft recommends video or in‑person interviews, pointing out that deep‑fake visuals often betray themselves through pixelation at the edges of faces, irregular lighting, or mismatched eye and ear geometry. It also advises recruiters to verify identity documents against known databases and to be skeptical of applicants who can instantly produce code or documentation without a verifiable work history. By highlighting these “tells,” the company hopes to curb the flow of AI‑enhanced fraud into the remote‑work ecosystem.

The broader implication is a new frontier for state‑sponsored cyber‑espionage: generative AI lowers the barrier to creating believable personas at scale, turning what was once a niche social‑engineering trick into a mass‑deployment recruitment weapon. As Microsoft’s own AI initiatives push toward “experts teaching machines,” the same tools are being co‑opted by adversaries to amplify deception, underscoring the dual‑use dilemma that policymakers and enterprises must now grapple with.