钢印 - 科幻短篇小说 | The Steel Seal — A Sci-Fi Short Story

2026-06-15 | WDSEGA

中文

一

在”绝对防伪”公司工作了十二年之后，老周第一次拒绝签发一枚钢印。

那枚钢印的申请来自省美术馆，认证对象是一幅据说是清代仿制的《兰亭序》。AI鉴定的结论是：71%真，29%不确定。71%不是”是”，29%也不是”不是”——但钢印就是钢印，要么盖要么不盖。

老周拒绝的理由不是原则，是因为那天早上他在AI鉴定系统的运行日志里看到了一行字——”参数权重校准已激活——监管指导模块 v2.4。”

一行没有人读的字。但老周是读编译警告的人。

二

老周辞职那天晚上复印了三份文件，分别交给了三个人：《南方周末》一个记者、鉴定部一个前同事、还有一个叫杜辰的人——河北一个造假画的。

文件是一份内部鉴定报告。一幅上拍的宋画，AI鉴定颜料成分与元代真品82%匹配，分析师建议”可以认定真迹”，第三方国际实验室已经确认。然后——认证科给它盖了一个”存疑”。

没有解释。不是推翻分析师结论，也不是质疑国际实验室的结论。就是”存疑”两个字。

老周查了这幅画的拍卖记录——拍卖会安排在这份鉴定报告出具后的第四个月。起拍价一千二百万。”存疑”两个字打下去，估价调整到两百万。如果有人在存疑之后重新走一遍认证——低价买入，重新认证，高价卖出——两百万和一千二百万之间，差价一千万。什么都不用改，两个字。

老周不确定这件事是否真的发生了。他只是知道：这件事现在可以发生了。一旦一件事”可以”发生，它就一定会发生，只是时间问题。

三

记者想立刻发报道，前同事想匿名举报，杜辰——造假画的那个——只是看着老周递给他的信封，不说话。

他在一个下雨的晚上敲了老周的临时住处的门，手里还拿着那个没拆开的信封。

“你知道，”杜辰说，”我造的三百多幅假画，和别人的假画有什么区别？”

老周看着他不说话。

“每一幅假画的背面，我用炭笔写了一个’杜’字。写在画心背面，扫描仪看不见，X光也看不见。一两百年以后的人才看得见——因为炭笔会缓慢氧化，留下一个不可逆的印子。”

他说：”我学了二十年画画。我不是把别人的名字签在我的画上，我是把自己的名字签在别人的画上。有区别。”

四

老周坐了六个小时的绿皮火车的硬座，去了一个中部省份下辖的一个县城，投靠他爷爷最后一个徒弟——一个叫曹瘸子的人。

曹瘸子年轻的时候跟着老周的爷爷学了六年铁章。后来他没干这一行，改养猪了。他住在一间红瓦房的尽头，猪圈在后院，灶台在堂屋，床底下压着一个木箱子。

木箱子里有四枚铁章——老周爷爷的遗物。每一枚一个汉字：真、伪、存、疑。

曹瘸子拿出一枚，放在桌上。”这是最后刻的。你爷爷刻完这枚章三天后走的。”章的正面刻了一个字——”真”。

老周把章拿在手里，很重。比他公司那个嵌了AI芯片的电子钢印重得多。那枚电子钢印的运作原理是这样的：激光扫描→光谱分析→分子层AI判定→全自动盖印。全程没有人干预。老周在公司做了十二年，他这个”分析师”的职位的真实作用不是分析，是在AI判定完之后复核——但AI判定完之后，他已经不需要复核了。他的工资是为”有人在看”这句话买的单。

他忽然明白了那行日志是什么意思：”参数权重校准已激活——监管指导模块 v2.4。”

五

曹瘸子说：你帮我刻个”真”字。

老周坐在那间屋子里，用一把刻刀、一截生铁、一盏台灯，花了四个小时二十七分钟，刻出了他这辈子第一枚手工章。他不确定那枚章能不能用，只确定一件事：这枚章的参数权重，没有人可以远程校准。

第二天他回了公司，把那枚铁章放在前台，说出了那句他在心里存了七天的话：”如果一枚章的真假需要用另一枚章来认证，我们就陷入了无限递归。递归的终点，必须是一个不能被数字修改的东西。”

前台姑娘没听懂。

老周也没指望她听懂。

English

I

After twelve years at “Absolute Anti-Counterfeit” Corporation, Old Zhou refused to stamp a certification for the first time.

The AI had rated the painting 71% genuine, 29% uncertain. Not a yes, not a no. But a stamp is binary — stamped or not.

Zhou refused not on principle. He refused because that morning, in the AI system’s runtime log, he’d seen an auto-generated line: “Parameter weight recalibration active — regulatory guidance module v2.4.” The kind of line no one reads. But Zhou was the kind of person who reads compiler warnings.

II

The night he resigned, Zhou copied an internal authentication report three times and handed it to three people. The report concerned a Song Dynasty painting scheduled for auction. AI analysis matched the pigments 82% to authentic Yuan-era materials. The analyst recommended “genuine.” A third-party international lab confirmed. Then — the certification office stamped “Uncertain” over it all.

No explanation. Just two characters.

Zhou checked the auction calendar. Four months after the stamp. Starting price: ¥12 million. After “Uncertain,” revised estimate: ¥2 million. A ¥10 million gap between two characters. Nothing needed to change except who held the stamp.

Zhou didn’t know if it happened. He knew it could. Once a thing can happen, it will.

III

The third recipient was Du Chen, a forger from Hebei. He showed up at Zhou’s temporary apartment on a rainy night, envelope still unopened.

“You know the difference between my 300 forgeries and everyone else’s?” Du Chen asked. “On the back of every piece I forged, I wrote my name. Charcoal. Invisible to scanners. A hundred years from now, someone will see it. Charcoal oxidizes.”

He’d studied painting for twenty years. “I’m not signing someone else’s name on my work,” he said. “I’m signing mine on someone else’s work. There’s a difference.”

IV

Zhou took a six-hour train to a county in central China, to find his grandfather’s last apprentice — a man called Cao the Cripple who’d given up iron seals to raise pigs.

Under Cao’s bed was a wooden box holding four iron seals, each carved with a single character: Real. Fake. Doubtful. Uncertain.

Cao lifted one out. “This was the last one. Your grandfather died three days after carving it.” The character carved into the face: “Real.”

Zhou held it. Heavy. Heavier than the AI-chip-embedded electronic stamp at his company. That stamp worked by: laser scan → spectral analysis → molecular AI verdict → auto-stamp. No human intervention. Zhou’s job title was “analyst” but his real function was being the person who confirmed the AI’s decision after it was already made. His salary bought one thing: the appearance of human oversight.

Suddenly he understood the log line: “Parameter weight recalibration — regulatory guidance module v2.4.”

V

Cao said: “Carve a ‘Real’ for me.”

Zhou sat in that room with a carving knife, a piece of raw iron, and a desk lamp. Four hours and twenty-seven minutes later, he had carved his first handmade seal. He wasn’t sure it would work. He was sure of one thing: no one could remotely recalibrate this seal’s parameter weights.

The next day, he walked into Absolute Anti-Counterfeit and placed the iron seal on the front desk.

“If every stamp needs another stamp to verify it,” he said, “we’re in infinite recursion. The recursion has to bottom out at something that can’t be modified by a number.”

The receptionist didn’t understand.

Zhou didn’t expect her to.