The Employee Quit. Her AI Clone Kept Working. The Take Inside China

A worker resigns: desk cleared, building access revoked, work-chat groups exited. But “the other self” stays — an AI digital human trained on years of chat logs, work documents and project materials, answering clients in her tone, handling business with her experience, writing status reports in her style. Our WeChat column’s take (2026-08-30, “You can resign; your digital clone can’t”) opens on that image and argues it stopped being science fiction in April 2026. This entry translates the take with the April record pinned beneath it — including the one claim that did not survive checking.

The numbers

• April 2026 coverage tier: Dahe Daily’s original interview (2026-04-06), CNR’s 看丹观察 (2026-04-09), Science & Technology Daily with named experts (2026-04-09), China.com.cn commentary (2026-04-15), Caixin Weekly (2026-04-25), Xinhua (2026-04-27)
• Weibo: #公司用AI复刻离职员工继续工作# peaked #1 on 2026-04-07, heat ~1.066M, on-list 1h34m
• The enabler: colleague.skill, open-sourced 2026-03-30, ~7,000 GitHub stars in 5 days; ingests Feishu/DingTalk chats, docs, email, screenshots
• The case: Shandong game-media firm (~100+ staff), clone of a departed HR specialist — queries, invitations, slides, spreadsheets; internal beta; consent claimed by the posting coworker
• Legal exposure cited: PIPL + Interim Generative-AI Measures; infringing citizens’ personal information — up to 3 years, 3–7 if serious
• English coverage: SCMP, ThinkChina, MIT Technology Review (2026-04-20)
• Figure arbitration: the take’s “recently re-trended on hot search” — no August re-trend found; the moment was April. Two of its source attributions are wrong: the “digital workers” worry-piece is China.com.cn (not Economic Daily), and the “serious issue needing clarification” commentary is Chuanguan News (Sina republished)

The take inside China

Why companies do it. From the employer’s seat the logic is nearly perfect: what hurts when a veteran leaves is not the empty desk but the experience that walks out — the client’s temperament mapped, the project landmines stepped on, the calibrated tone of every conversation — none of it written into any handover document, none of it reproducible in a year or two of training a replacement. Then AI makes the offer: give me the chat logs and documents, and I will keep that experience. Cost reduction and efficiency, in their most concrete form ever.

Where the dread sits. The top-voted comment was “terrifying on reflection” — and the take dissects the reflection into three layers. First: whose is my experience? Work documents as company assets is uncontested; but my speaking style, my communication habits, the judgment forged by stepping in pits — things that grew on my person — replicated into a model that belongs to the company? I can’t take it with me; the company keeps it forever. Second: the authorization is fictional. Even if the employment contract said “the company may use work data,” could the person who signed ten years ago have anticipated that “use” includes cloning them? The clause didn’t change; the technology changed its meaning. Third: the clone speaks as me — messages to clients in my voice, emails under my name, recipients believing it’s me. Your professional personality is borrowed, with zero control over what it says; when it errs, whose reputation does it burn? The take’s thesis: this is not a labor dispute but a new proposition of the digital era — for the first time, the commercial part of a personality can be left behind at the employer. The last generation surrendered a badge and settled accounts; this generation may be asked to surrender its digital shadow.

The skeptics get their paragraph. Companies have always retained experience — knowledge bases, retrospectives, mentorship; the digital human is an old demand in new packaging, and today’s “digital employees” are more gimmick than substitute for judgment. Both fair, the take concedes — but the distinction is that what was retained before was knowledge, and what is replicated now is the person-flavor. Knowledge in documents is neutral; tone and style living inside a model are an extension of personality — and personality was never company property.

Before the law catches up. Read the data-use clauses at entry and exit for the words “training,” “model,” “digital retention.” Decide what lives in company systems and what stays in your head. And if a clone is speaking in your name: preserve evidence, get a lawyer — the legal blank here is exactly the space the next few years of legislation and litigation will fill, and the first people who take it seriously will shape the rules.

What the Chinese take left out

The concrete case is one, company-anonymous, and resting on a pseudonymous coworker’s account of consent — no named company, no independent verification beyond April’s reporting. The August re-trend claim does not check out (the hot-search moment was April 7; the only later echo is a June 25 legal-analysis piece). Two source attributions in the take’s footer are mislabeled. And the take never mentions that English coverage existed early and long: MIT Technology Review profiled the project’s author and a user recreating an ex-coworker, and an anti-distillation counter-tool built in response drew over 5 million likes — the phenomenon had already traveled.

Why it matters outside

China has produced the sharpest legal instruments yet applied to post-employment cloning — a consent regime (PIPL), a generative-AI training rule, and a criminal statute the cited lawyers say applies — while the enabling tool went from release to 7,000 stars in five days. That combination (consumer-grade open-source vector, named criminal exposure, zero adjudicated cases) is the full question set for every jurisdiction: personality rights vs. work product, consent scope vs. secondary use, and who owns the tacit part of a person’s craft. The first ruling anywhere will be a global reference.

Sources

• CNR 看丹观察: departed employee trained into an AI digital human
• Dahe Daily (via Jimu News): the Shandong firm’s HR clone — original interview
• China.com.cn: why “digital workers” from departed employees worry us
• MIT Technology Review: inside China’s colleague.skill craze
• Weibo hot-search record: peaked #1 on 2026-04-07

Provenance & disclosure. Originally published in Chinese on our WeChat channel on 2026-08-30 (“人可以离职，你的’数字分身’不行”); drafted with AI assistance under human editorial direction. Translated to English on 2026-08-30 (AI-assisted, human-reviewed). This entry goes beyond translation: April’s coverage chain was re-verified outlet by outlet (with two attributions corrected), the case’s company-anonymous status confirmed, the hot-search record pinned to April 7, and the take’s August re-trend claim marked unverified after six searches found no trace. This is translated commentary — not a SigPulse measurement. Our first-party measurements live in the dispatches and the /data/ ledger.