Security1 distinct publisher3 min readUpdated
Recorded Future says one PurpleDelta cluster applied at more than 1,100 companies and was likely employed at ten or more. The control point is the recruiting funnel, not the SOC.
The Watch · Security desk

Compiled by The WatchSomething wrong?How this is made
Recorded Future's Insikt Group has published research on PurpleDelta, its designation for North Korean IT workers, describing operators likely based in China who applied for jobs at more than 1,100 companies between late 2024 and early 2025 [1][2]. Insikt assesses those clusters were highly likely to be actively employed at ten or more organizations [3], which locates the failure in the hiring funnel rather than in anything a detection team was ever going to see first.
The volumes describe a production line. Insikt counted at least 22 fabricated personas across multiple clusters, submitting as many as 60 or more applications per day across at least eight job platforms [4][5]. At that tempo, covering 1,100 companies is roughly 18 working days of output [1], and 1,100 companies spread across 22 personas averages about 50 targets each [2]. Ten placements against 1,100 targeted companies is a conversion rate near 0.9 percent [3]: unremarkable for a job seeker, and entirely sufficient when applications cost nothing and the persona inventory never runs dry.
The tooling is pointed at recruiters, not at firewalls. Personas were supported by AI-generated profile photos, custom-configured ChatGPT assistants, and identity documents sourced from an illicit ID-generation service, according to Insikt [6]. Operators ran multi-account management browsers and separate Chrome profiles to keep identities apart, with tracking spreadsheets to coordinate applications across them [7]. In interviews they used screen recording software alongside AI transcription and chatbot tools to generate answers in real time, sometimes repeating ChatGPT output verbatim [8]. Verbatim ChatGPT is a detectable interview artifact, but only if a human is on the call and listening for it.
After the offer, the tells become administrative. Insikt reports that once employed, operators recorded internal meetings at victim organizations and drafted pre-written Google Translate excuses to justify using personal devices and personal bank accounts for work [9]. At least two individuals were identified as facilitators who procured and maintained company-issued hardware on the operators' behalf [10], and coordination ran over Telegram and Slack [11]. Those are payroll, asset-management and IT-logistics signals: laptops shipped to an address that is not the employee's, bank detail changes, a device that never appears from the claimed location.
The targeting is also worth reading as a sector map. Insikt's executive summary lists software and technology, staffing and consulting, and healthcare and biotechnology as the primary sectors [12]; its key findings add financial services [13]. Staffing and consulting is the entry with leverage against it, because a placement there sits next to other companies' pipelines.
PurpleDelta overlaps with designations other vendors use, including Jasper Sleet, UNC5267, Wagemole, and Famous Chollima [14]. Earnings are funneled through layers of individual facilitators, shell companies, and money-laundering front companies to finance North Korea's sanctioned military and nuclear programs, per Recorded Future [15].
What to watch: Insikt says companies that have observed the indicators in its Appendix A should treat the situation as a potential active compromise and review the employment history and access privileges of matching individuals [16]. That is a records exercise owned by HR, talent operations and IT asset management, and it looks backwards at people already onboarded rather than forwards at candidates. Insikt also notes the operators demonstrate a high operational tempo to this day [17], so any control built around a single interview round is being tested continuously by a pipeline that only needs 0.9 percent to work.
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
Between late 2024 and early 2025, one PurpleDelta cluster applied to jobs at over 1,100 companies.
Insikt Group identified at least 22 fabricated personas linked to multiple PurpleDelta clusters that submitted applications to over 1,100 companies.
Operators submitted as many as 60 or more applications per day across at least 8 job platforms.
Insikt Group has identified several clusters of activity linked to PurpleDelta, Recorded Future's designation for North Korean IT workers, comprising multiple operators likely based in China.
The PurpleDelta clusters were highly likely to have been actively employed at ten or more organizations, with confirmed or probable placements posing an ongoing and material insider threat.
Some PurpleDelta personas were supported by AI-generated profile photos, custom-configured ChatGPT assistants, and identity documents sourced from an illicit ID-generation service.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Detailed first-party forensics, single vendor, no external corroboration
The report supplies specific, checkable quantities (1,100-plus companies, 22 personas, 60-plus applications/day, at least 8 platforms, ~41/26/10% sector split, ~80% North America) and says findings derive from recorded sessions and video evidence, which is unusually concrete for this threat class. It is nonetheless one publisher's own research: no victim organizations, law-enforcement action, or independent replication is cited, key placement findings are hedged as 'highly likely' and 'confirmed or probable', and the target-sector list is stated inconsistently between sections. Appendix A indicators are referenced but not shown in the supplied text.
Real placements asserted at scale, but unnamed and unverifiable externally
Adoption here is the attacker-side operational footprint, and it is substantial on the source's own account: a documented 1,100-plus company funnel, ten or more organizations assessed as actively staffed by operators, at least two hardware facilitators, and a full commodity AI and multi-accounting toolchain in production use. It is held below the midpoint plus because no employer is identified, no placement duration or access level is given, no post-early-2025 volume figures back the 'to this day' framing, and remediation uptake by employers is not observed at all.
Mildly overstated framing around a well-quantified core
The numbers themselves are presented carefully and the arithmetic holds: roughly 18 working days of output at the stated tempo, about 50 companies per persona, and a conversion rate near 0.9%. The overstatement is at the edges rather than the center: a new vendor designation is layered onto activity already tracked as Jasper Sleet, UNC5267, Wagemole and Famous Chollima; 'high operational tempo to this day' is asserted without current-period data; and the sanctioned military and nuclear financing chain is stated as established fact without evidence specific to these clusters. The most alarming figure, 1,100-plus companies, describes applications rather than compromises.
Vendor research promoting its own actor designation and indicator feed
The sole source is the research arm of a commercial threat-intelligence provider publishing under its own actor name, PurpleDelta, and directing readers to its Appendix A indicators with instructions to treat matches as active compromise. That is a direct path from finding to product relevance for hiring, insider-risk and intelligence-feed offerings. The incentive does not by itself weaken the forensic detail, and the report is explicit about which findings are assessments rather than confirmations, but no non-commercial party is represented in the cluster.
Credible and specific, but wholly single-sourced with hedged core findings
Confidence is moderate: the tradecraft description is granular and internally coherent, and the volume math is consistent, which supports believing the operational picture. It is limited by having exactly one publisher, by hedged probabilistic language on the placements that matter most, by an unresolved internal discrepancy in the sector list, and by the absence of named victims, indicator content, or third-party confirmation that would let a reader test any figure independently.
security
Aeternum puts botnet C2 on Polygon, and leaves defenders no domain to seize1 distinct publisher
build
The AI tell is installed during fine-tuning. Humanizer tools edit the output instead.1 distinct publisher
security
Influence Operations Now Target Construction Schedules, Not Just Elections1 distinct publisher
invest
DeepSeek V4 Flash costs a tenth as much and passes 53.8% of agent tasks1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 17, 2026