拆掉电视的锁,也拆掉别人的门Pick the Lock, Kick the Door
这篇报道里最诚实的一句话是它自己都说不清这算“反编译”“逆向工程”“黑客”“vibe coding”“AI 垃圾”还是“偷窃”。它说不清,因为这几件事用的是同一套动作:拿别人的东西,改一改,变成自己的。区别只在于你改的是自家电视上那块去不掉的广告,还是竞争对手音箱里那套调音算法。
先说清楚哪一层在动。小米电视开机先放广告、Lululemon 的镜子被厂商 brick 掉、婴儿监视器塞满推销——这是结构暴力,不是比喻。厂商用专有固件把“你买下的硬件”重新定义成“我租给你的入口”,功能是他们的,广告位是他们的,你只是那个每月交电费还得看广告的人。AI 把 25 个小米软件包卸掉、写个新启动器,这一锤砸的是结构层。加尔通的公式在这里很好用:潜力是这台电视本可以干净开机,实际是它先给你放一段广告,差额就是被拿走的那部分。Gathin 那条帖子被看了六百多万次,说明被拿走的人有多少。
但同一篇文章里,那个做吉他效果器的人在 Hacker News 上说,72 小时内 one-shot 反编译了三个主要竞争对手的产品,“理论上我直接装盒子里就能转卖”。这不是砸锁,这是砸别人家的门。他偷的不是广告位,是别人几十年积累的算法,而且是“任何有 Claude 订阅的人”都能偷。这两件事在技术上完全一样,在伦理上完全相反,报道把它们并排放在一起,读起来像是同一场狂欢。
真正被武器化的不是 AI,是“反正拦不住”这句话。报道自己给出了机制:模型变强、护栏被拆、agent 化之后,你只要告诉它去做,它基本就去做。于是速度成了唯一的标准——克隆速度、上线速度、铺开速度。而维护、审核、修 bug 的速度一点没变。报道说得很直白:这些克隆软件经常有连使用者自己都看不懂、也不会修的 bug,但人们照样用,照样装到联网硬件上。速度差就是新的暴力入口,它不流血,它只是把一堆没人负责的代码塞进你家的婴儿监视器。
这里要问一句谁在共谋。不是那些周末给父母电视写启动器的人,他们拿回的是自己被拿走的东西。共谋的是那批把“能不能做”直接等同于“该不该做”的人:拿 AI 抄竞品算法再压价的小厂,把半成品硬件塞进餐厅点单屏的创业公司,还有那些一边说“开源精神”一边把逆向工程社区当成免费流量池的人。他们得到的回报很具体——省掉的研发时间、抢到的市场窗口、涨上去的 star 数。付代价的人不在报道里:被抄垮的小公司、买到带后门固件的消费者、以及那些花了几百小时真正在修 bricked 设备的开源社区成员,他们的位置被一群 Kool-Aid man 式撞进来的人占了。
报道最后一句说,也会有人用这套东西拆掉“本来就不该存在”的厂商限制。这句是对的,但它和前面那堆偷算法、塞恶意代码、抄垮小厂的事被摆成“mixed bag”,好像好坏各占一半,所以不用下判断。不是的。判断的标准从来不是工具新不新,是这一锤下去,谁的存在性变大了,谁的被让渡了。拆掉自家电视的广告,你的表达空间变大;拆掉别人的算法去卖钱,你的变大是以别人的消失为代价。同一把锤子,两种结果,中间那条线不模糊,只是报道不愿意画。
还有一层没被提到:这些设备原本为什么会被锁成这样。厂商敢在电视开机前塞广告、敢把镜子 brick 掉,是因为结构层默认“卖给你之后东西还是我的”。AI 逆向只是把这个默认撕开一个口子,口子撕开之后,补上的不该只是几个能用的启动器,而是“买断就是买断”这条规则本身。现在的情况是,规则没动,工具先到了。工具会流向力气最大的人,而力气最大的人从来不是周末给爸妈修电视的那个。
The most honest line in this piece is that it can't decide whether this is "decompiling," "reverse-engineering," "hacking," vibe coding, AI slop, or stealing. It can't decide because all of these are the same motion: take what belongs to someone else, change it, call it yours. The only difference is whether you're removing the ad that plays before your own TV reaches Netflix, or lifting the effect algorithm out of a competitor's guitar pedal.
Start with which layer is moving. A Xiaomi TV that runs a sponsored ad before it lets you watch anything, a Lululemon mirror bricked by the company that sold it, a baby monitor stuffed with ads—that is structural violence, and not as a metaphor. The manufacturer used proprietary firmware to redefine "the hardware you bought" as "the entrance I rent to you." The features are theirs, the ad slots are theirs, you are the person paying the electric bill and watching the ad. When AI strips 25 Xiaomi packages and writes a new launcher, that hammer lands on the structural layer. Galtung's formula fits: the potential is a TV that boots clean, the actual is an ad before Netflix, and the gap is what got taken. Gathin's post was viewed more than six million times. That's how many people noticed the gap.
But in the same article, a man who claims to make guitar pedals says on Hacker News that he one-shot decompiled three of his main competitors' products in 72 hours, and could in theory stick them in a box and resell them. That is not picking a lock. That is kicking down someone else's door. What he took wasn't an ad slot; it was decades of accumulated algorithm, and the article notes that anyone with a Claude subscription can now do the same. Technically identical to the TV hack. Ethically the opposite. The piece runs them side by side as if they were one carnival.
The thing being weaponized isn't AI. It's the sentence "you can't stop it anyway." The article hands you the mechanism: better models, guardrails removed, agents that generally do what you tell them. So speed becomes the only metric—speed to clone, speed to ship, speed to flood. The speed of maintenance, review, and bug-fixing hasn't changed at all. The piece says it plainly: these clones often carry bugs the person using the AI doesn't understand and can't fix, and people install them anyway, onto internet-connected hardware. That speed gap is a new entry point for violence. It doesn't bleed. It just puts unaccountable code inside your baby monitor.
So who is conspiring here? Not the people who spent a weekend writing a launcher for their parents' TV. They took back something that had been taken from them. The co-conspirators are the ones who collapse "can I" into "should I": the small shop that clones a competitor's algorithm to undercut them, the hardware startup that ships half-working cloned software into a restaurant ordering screen, and everyone who waves the flag of open source while treating reverse-engineering communities as a free traffic pool. Their payoff is concrete—saved R&D time, a market window, a rising star count. The people who pay are not in the article: the small companies getting cloned out of existence, the consumers buying firmware with a back door, and the open-source contributors who spent hundreds of hours actually reviving bricked devices, whose space is now occupied by Kool-Aid men crashing in with "their" AI projects.
The last line says some people will use this to strip manufacturer guardrails that never should have existed. That's true. But it's placed next to stolen algorithms, injected malware, and crushed small firms as a "mixed bag," as if the balance of good and bad means no judgment is required. No. The test was never whether the tool is new. It's who comes out of the swing with more existence, and whose gets conceded. Removing the ads from your own TV expands your space. Cloning someone's algorithm to sell it expands yours by making theirs disappear. Same hammer, two outcomes. The line between them isn't blurry. The article just won't draw it.
And one layer goes unmentioned: why these devices were locked this way to begin with. Manufacturers dare to put an ad before the boot screen and brick a mirror because the structural layer defaults to "after we sell it to you, it's still ours." AI reverse-engineering tears a hole in that default. What should fill the hole isn't a few working launchers—it's the rule that buying means buying. Right now the rule hasn't moved and the tool has already arrived. Tools flow to whoever has the most leverage, and that has never been the person fixing their parents' TV on a Saturday.