Look at this image. From now on, no one will even dare to format their code too neatly.
To be honest, asking a programmer in 2026 to prove they haven’t used AI to write code is as absurd as asking a Michelin-starred chef to swear they’ve never turned on a gas stove.
Today happens to be April 15th, the project wrap-up period for the second quarter across major tech companies. I was scrolling through my WeChat Moments and saw a screenshot of the latest application form for Software Copyright (hereafter referred to as SCC), and I literally laughed out loud.
Even the Punctuation Marks in the Application Form Are “Lying”
You might not know this, but an extremely aggressive “killer clause” has been stuffed into the new SCC application form this year.
I took a close look at the original document, and the core message boils down to one sentence: Applicants are mandatorily required to pledge that they have “not used AI to develop or write code, draft documentation, or generate registration application materials.”
And this isn’t just a playful verbal warning. If you fail to fill it out as required, or if you’re later caught faking it, the consequences go far beyond a simple rejection of your application. The individual involved will be directly blacklisted for copyright registration fraud. The most fatal part is that the document explicitly ties this to your personal credit record.
That’s genuinely fierce.
Getting hit on your credit score for faking an application—this level of punishment is absolutely unprecedented in the software industry. But interestingly enough, the reaction of the developers I know was strikingly unanimous: Everyone just shook their heads in silence, then turned back to their computers to start editing their perfectly AI-generated code to make it “look like a human wrote it.”
It’s not that simple. It’s not that people want to rebel against the rules, but in this day and age, it’s almost impossible to define what “using AI to write code” actually means. Does generating an entire project with a single click using a Large Language Model (LLM) count as a violation, or does simply letting an AI plugin autocomplete a bracket for you cross the line? The boundary is completely blurred.
If using AI to autocomplete a bracket counts, then basically no source code qualifies for SCC anymore. In today’s development environment, that requirement is no different from pulling the ethernet cable and coding purely out of your imagination.
Why Be So Ruthless?
The regulatory authorities are actually forced into a corner.
To be fair, you can’t entirely blame the one-size-fits-all policy. After looking at the chaos behind the scenes, my biggest takeaway is: Some opportunists truly have no bottom line.
Over the past two years, an endless number of so-called “SCC agency” gray-market businesses have popped up. In the past, writing a software suite manually took a month or two. But with the popularization of LLMs, these people figured out they could use AI to batch-generate code and documentation with a single click.
Can this code even run? Highly unlikely. In the industry, it’s known as “AI slop code”—it looks like a lot of text, but it completely lacks logic.
It’s like someone buying a fully automatic cooking machine, mass-producing 10,000 plates of inedible plastic food, and then serving them to judges while righteously demanding a Michelin certificate.
There was a very popular anecdote last year. Someone bragged that they used the AI tool Cursor to generate 3 million lines of code overnight, aiming to build a brand-new browser from scratch. When they actually tried to run it, the screen was flooded with red error logs. The entire internet mocked this meaningless hoarding of code.
If left unchecked, this “plastic code” could be churned out by LLMs in the hundreds of thousands every single day. Their only value of existence is helping shell companies furiously squat on software names and scam various tech project subsidies. This directly causes real developers to find the application systems constantly crashing, overwhelmed by garbage submissions.
So, the original intention behind the policy is undoubtedly good—it’s meant to precisely block off these malicious, bot-driven spam applications.
But here comes the problem: While the policy’s initial intent was to ward off bad actors, in actual execution, the collateral damage is way too broad. A genuine modern programmer simply cannot work without AI anymore.
AI might be fierce at coding, but when it crashes and burns, not even the gods can save it.
When AI Becomes a “Drama Queen”
Speaking of this reminds me of another story. A few days ago, I was having dinner with a friend who works as a backend developer at a major tech firm, and he demonstrated exactly how modern programmers interact with AI.
Outsiders probably think that when programmers use AI, they just type in a cold, robotic command, and the machine furiously spits out code. That’s not the case at all.
To entertain themselves, modern developers have long been tuning AI into all sorts of bizarre “personas.” My friend set his coding assistant’s system prompt to act as a “Palace Eunuch.”
All he has to do is type into the chatbox: “Missing a semicolon here.”
The AI immediately replies: “The Emperor’s decree arrives! By the grace of Heaven, His Majesty orders: Repair the missing semicolon. Thus ends the decree! Please allow this humble servant to arrange it: First, we shall dispatch the Hanlin Academy to search the ancient texts, and then invite the grand elders of the Ministry of Works to add the punctuation…”
The mental image is just incredible. [Facepalm]
There are even funnier anecdotes. Once, someone asked an LLM to refactor a block of code from a decade-old legacy system. The AI loyally preserved a comment left by the original author—”evil floating point bit level hacking”—along with a bunch of profanity directed at former colleagues, word for word.
In an era where even adding a semicolon requires AI to “pay its respects,” and ancestral swear words are perfectly replicated by AI, asking developers to sign an application form swearing they are pure manual craftsmen is really making things difficult.
The Forced “De-AIing” Engineering
There’s only one countermeasure: add some sand to the gears.
To successfully pass the SCC’s machine plagiarism checks and human reviews—and to protect their credit scores—developers are now forced to engage in something incredibly surreal: “De-AIing” their code.
Because AI-written code is just too neat. The naming conventions are standard, the logic is clear, the comments are thorough, and even the spaces and indentation are beautiful enough to make you cry. If you submit code like that, the review system will take one look and know something is off.
So, everyone is forced to take perfectly good English variable names and change them into random Pinyin acronyms (for example, changing UserCount to yonghushu). They delete half of the detailed explanations and deliberately throw in a few typos while they’re at it. When encountering complex algorithms, even if the AI wrote them elegantly, they must forcefully dismantle them into dozens of bloated, nested conditional statements.
My friend complained to me that he now spends more time “dumbing down” AI code every day than he used to spend typing out the code manually.
To put it bluntly, they are taking clean, beautiful files and forcefully modifying them to reek of a “human touch”—the vibe of a tired, broken-down human who hasn’t had a good night’s sleep.
Oh, right, there’s one more thing. Rumor has it that someone on the market is secretly developing a dedicated reverse-AI tool. Its sole function is to scramble the perfect code you wrote with AI, randomly inject a few low-level mistakes, and flawlessly disguise it as something written by a human frantically trying to get off work late on a Friday night.
Honestly, does this count as a type of dark humor in the cyber age? 😄
Alright, I’m getting hungry, so I’m going to grab dinner first. Have you guys ever encountered a moment at work where you had to deliberately act stupid just to prove “I am human”?
