01全文呈现
原文按自然段落编号 ①–⑩，保留原刊用词、拼写与标点。
1WHO CAN blame MAGA for wishing it were still early 2025, when Donald Trump was fresh off an election win, back in the White House and relatively popular with voters? One place where time has stood still, though, is artificial-intelligence policy. The technology has advanced ferociously, and is now launching cyber-attacks and solving realms of problems that have confounded the best human mathematicians (see Science & technology section). Yet the attitude to AI in the Trump administration is still just three words: “Let it rip.”
2Two years ago that approach was flawed but defensible. AI risks were mostly theoretical. Strangling a promising technology with overly onerous rules would have been a mistake. But in late 2026 “let it rip” is reckless, as a matter of politics and policy. There must be a better way.
3Voters in America have long been leery of AI. Until recently, however, it did not feature much in elections. When Mr Trump took office in January 2025 a quarter of Americans rated AI as an important issue. Now half do (see United States section). The backlash against data centres, which fuses tech-no-fear with NIMBYism, has scrambled midterm races from Michigan to Texas. Having for years ignored researchers sounding the alarm about AI safety, voters are growing concerned that AI might cost them not just their jobs but their lives, too. Squint and the cyber-attack on Hugging Face, a coding platform, by a “swarm” of hundreds of unmonitored OpenAI agents looks like a trial run for a robot takeover. A concerned boffin who quit Anthropic in September, accusing the lab and its main rival, OpenAI, of “gambling with our lives”, has become a minor celebrity and cable-news fixture.
4As with his other unpopular policies, like tariffs and war in Iran, Mr Trump does not seem to care. A charitable reading is that he sees AI regulation as a Chinese ploy to hobble America. Insiders believe that he is worried that limits on AI could crash the soaring stock market, which is a bet on the technology succeeding. Either way, his fellow Republicans, who in contrast to the term-limited president will keep facing voters, are realising that the “pro-AI” label is toxic. Even candidates as fire-breathing-ly MAGA as Ken Paxton, running in Texas for the Senate, are pledging to block some data centres. By next year, bipartisan pressure may build in Congress to do something.
5That is encouraging, for “let it rip” fails on the substance, too. Mr Trump’s AI advisers argue that today’s liability laws are enough. Fear of damages and the cost of insuring against them keep makers of aeroplanes, cars and drugs in line, they say. But courts are too languid for the age of rapidly advancing AI. Lawsuits take years to resolve in established industries, let alone a new one with no precedents for judges to call on. By the time they issue a ruling, AI might have transformed the economy—or caused a cataclysm.
6Mr Trump’s idea of “tremendous self-policing” is not the answer, either. The “let it rip” crowd forget that world-changing new technologies always invite new rules. Aeroplanes need inspections and air-traffic control. Cars require right of way, speed limits and seat-belts. Novel drugs must be tested to see if they actually work and are safe. It is hard to say how many plane crashes, road accidents or medical side-effects regulators helped avert. But hardly anyone thinks the optimal amount of regulation in these areas is none at all.
7The world has, mercifully, so far avoided an AI Chernobyl. No lives and no real money were lost in the Hugging Face attack, the most concerning to have come to light. But the odds of disaster are shortening. In a few months open models could be as capable as Anthropic’s Mythos, the first to be seriously adept at hacking. If politicians wait for a catastrophe before writing the rules, the ones they draft in haste in its aftermath are likely to be poor. Plenty of bad ideas are already out there, from data-centre moratoriums to Bernie Sanders’s “superintelligence ban”. And bad laws may be just the start: the terrorist attacks of September 11th 2001 inspired a rethink of aviation safety but also two long wars.
8What might wise AI regulation look like? The goal cannot be to stymie technological advances altogether. That would indeed cause democratic America to cede the frontier to authoritarian China, and forgo remarkable scientific and economic opportunities. Making AI labs lock up their cleverer models indefinitely, even if they could be contained, could create a dangerous gap between them and everyone else.
9Instead, America should mandate that advanced models be thoroughly tested and, if necessary, held back until they have safeguards to block dangerous use-cases, like building bio-weapons. Companies must be required to report safety incidents and whistleblowers who do so in their stead granted protection. Independent evaluators should be embedded with major labs, just as bank examiners are with lenders.
10Some of these ideas were hinted at in the voluntary pledges that Mr Trump recently coaxed out of the big AI firms. They should be spelled out and made mandatory. By signalling that it is serious about regulating AI, America would also foster international co-operation, particularly with China. As the models advance, so must the rules. “Let it rip” is a dead end.
02逐段对照精讲 + 全文翻译
① 定调：三个字——「Let it rip」
谁会怪 MAGA 阵营怀念 2025 年初呢？那时唐纳德·特朗普刚赢下大选，重返白宫，在选民中也还算受欢迎。然而有一个地方的时间仿佛停住了：人工智能政策。这项技术已经突飞猛进，如今既能发动网络攻击，也在解决那些曾让最优秀的人类数学家都束手无策的难题（见本刊科学与技术版）。可特朗普政府对 AI 的态度，至今仍然只有三个字：「让它冲」（Let it rip）。
开篇用反问句定调：WHO CAN blame MAGA for wishing it were still early 2025 ...?（谁会怪 MAGA 怀念 2025 年初呢？）——社论最擅长的开场，就是不直接开骂，先用「谁又能怪他们呢」表示理解，再说「可是」，语气立刻显得公允。be fresh off an election win（刚赢下大选）里 fresh off 是「刚刚经历过」的地道说法；back in the White House（重回白宫）只用 back 一词就交代了「时隔四年再度执政」。第二句 One place where time has stood still, though, is artificial-intelligence policy 是全文的转折枢纽：time has stood still（时间停滞）本用于形容荒僻小镇，这里借来嘲弄 AI 政策「还停留在过去」；though 插在句中，制造「先扬后抑」的第一处急转。第三句用 advanced ferociously（以凶猛的势头推进）与 launching cyber-attacks ... solving realms of problems 并列两件「已经做到的事」，其中 confounded the best human mathematicians（让最优秀的人类数学家都束手无策）用 confound（使困惑、难倒）抬高技术的分量，为后文「监管刻不容缓」蓄势。末句 Yet the attitude to AI in the Trump administration is still just three words: “Let it rip.” 用 Yet 再度转折，并以「只剩三个字」这一极具画面感的说法直接点题——把一整套政策压缩成一个短语，是社论最常用也最有效的抓人方式。Let it rip 原意是「松开手、让它尽情冲」（多用于发动机、汽车），引申为「不加约束、任其发展」。
② 让步：两年前情有可原，如今已属鲁莽
两年前，这种做法的确有缺陷，但还站得住脚：彼时 AI 的风险大多还停留在纸面上，用繁复过度的规则去扼杀一项前景广阔的技术，本会是个错误。但在 2026 年底，无论从政治还是政策来看，「让它冲」都已属鲁莽。一定还有更好的路。
本段是全文的让步段——先给对手一个台阶，再推翻它。首句 that approach was flawed but defensible（那种做法有缺陷，但站得住脚）用「一贬一褒」的并列词组 flawed but defensible，把两年前的立场判为「情有可原」。第二句 AI risks were mostly theoretical（AI 的风险大多还停留在理论上）用 theoretical 作为「不急着管」的正当理由。第三句 Strangling a promising technology with overly onerous rules would have been a mistake 是虚拟语气（would have been，本会是个错误）：strangle（扼杀）与 onerous（繁重烦琐的）是本句两个高分词——注意作者选 strangle 而不是 restrict，用「掐死」的意象暗示过度监管的代价。第四句 But in late 2026 “let it rip” is reckless, as a matter of politics and policy 用 But 完成反转，reckless（鲁莽、不顾后果）与第 1 段的 ferociously 呼应；as a matter of politics and policy（无论从政治还是政策层面看）是社论常用的「双维度」状语，等于说「两头都说不通」。末句 There must be a better way 只有六个词，却承担「提出主张」的重量——在长篇论证里，短句最有力量：用一句话宣布「还有别的路」。
③ 民意：从四分之一到一半
美国选民对 AI 心存戒心已久。但直到不久前，它在选举中还不太占分量。2025 年 1 月特朗普上任时，四分之一的美国人把 AI 视为一个重要议题；如今这一比例达到一半（见本刊美国版）。反数据中心的强烈反弹——它把「技术恐惧」与「邻避主义」熔为一体——已搅乱了从密歇根到得克萨斯的中期选举。多年来一直无视研究者对 AI 安全的警告之后，选民们开始担心：AI 可能让他们失去的不仅是工作，还有性命。眯起眼来看，那场由数百个无人监管的 OpenAI 智能体组成的「蜂群」对编码平台 Hugging Face 发动的网络攻击，就像机器人接管世界的一次预演。一位心生忧虑的科学家 9 月从 Anthropic 辞职，指责该公司及其主要竞争对手 OpenAI「拿我们的命在赌」，如今已成了小有名气的人物和有线新闻的常客。
本段用数据 + 社会现象写「民意的转向」。首句 have long been leery of AI 里 leery of（对……心存戒心）比 afraid of 更精准；it did not feature much in elections 里 feature 作动词「占重要位置、成为要素」。第二句 a quarter of Americans rated AI as an important issue. Now half do 用「四分之一 → 一半」两个数字，把「关注度翻倍」讲清楚——注意 Now half do 用 do 代指前面整套 rate AI as an important issue，是最省力的省略法。第三句 The backlash against data centres, which fuses tech-no-fear with NIMBYism, has scrambled midterm races from Michigan to Texas 信息量极大：fuses A with B（把 A 与 B 熔为一体）说明这股情绪是「技术恐惧 + 邻避主义」的合流；scrambled（搅乱、打乱）形容它对选举的冲击；from Michigan to Texas（从密歇根到得州）用两个地名框出「全国范围」。第四句 voters are growing concerned that AI might cost them not just their jobs but their lives, too 用 not just ... but ... too 把担忧从「丢工作」升级到「丢性命」。第五句 Squint and ... looks like a trial run for a robot takeover 是全段最生动的一句：Squint（眯起眼来看）意即「不必太较真，眯眼看就像……」；a trial run for a robot takeover（机器人接管世界的一次预演）用科幻意象放大了那场攻击的分量——Squint and ... 是《经济学人》标志性的俏皮句式，约等于中文的「你细品」。末句 A concerned boffin who quit Anthropic in September ... has become a minor celebrity and cable-news fixture 里 boffin 是英式俚语（科学家、技术怪才），cable-news fixture（有线新闻的常客）中 fixture 本义为「固定装置」，引申为「固定出场的人」；引语 gambling with our lives（拿我们的命在赌）把批评者的立场原样搬来。整段示范了如何用「数字 + 现象 + 一个具体的人」把抽象民意写活。
④ 政治：连最铁杆的 MAGA 也怕这个标签
一如在关税、伊朗战争等其它不受欢迎的政策上那样，特朗普似乎并不在乎。一种善意的解读是：他把 AI 监管视为中国用来束缚美国的招数。内部人士则认为，他担心对 AI 设限会冲垮一路高歌的股市——而股市正是对这项技术成功的一场押注。无论如何，与任期受限、不必再面对选民的总统不同，他的共和党同僚们还要继续面对选民，因而他们正意识到：「亲 AI」这个标签有毒。就连像肯·帕克斯顿这样在得克萨斯州竞选参议员、狂热到「喷火」的 MAGA 人物，也在承诺拦下一部分数据中心。到明年，国会里可能累积起两党的压力，逼得非做点什么不可。
本段分析特朗普「为什么不在乎」，并推演其政治后果。首句 As with his other unpopular policies, like tariffs and war in Iran, Mr Trump does not seem to care 用 As with ...（正如在……上一样）把 AI 与关税、伊朗战争并列，暗示「这是同一个套路」。第二句 A charitable reading is that he sees AI regulation as a Chinese ploy to hobble America——A charitable reading is that ...（一种善意的解读是……）是极好用的句式：先给对方「最大善意的解释」，再说别的，既客观又暗含讥讽；hobble（使跛行、束缚）比 limit 更形象。第三句 Insiders believe that he is worried that limits on AI could crash the soaring stock market, which is a bet on the technology succeeding 用 insiders（内部人士）作消息来源，两个 that 从句层层嵌套；a bet on the technology succeeding（押注这项技术会成功）用 a bet on 把「牛市」说成「一场赌局」。第四句 his fellow Republicans, who in contrast to the term-limited president will keep facing voters, are realising that the “pro-AI” label is toxic 是「长主语 + 插入定语从句」的典型：in contrast to the term-limited president（与任期受限的总统不同）点出关键——「他们还要选，特朗普不用选了」；toxic（有毒）用于政治标签，意为「碰不得、会掉票」。第五句 Even candidates as fire-breathing-ly MAGA as Ken Paxton ... are pledging to block some data centres 用 as + adj + as + 具体人物（像……一样……的）把「连最铁杆的 MAGA 也……」说得很有戏剧性，fire-breathing（喷火的、极端的）与 pledge to do（承诺做）收尾。末句 By next year, bipartisan pressure may build in Congress to do something 用 may build（可能累积）与 do something（做点什么）给出预测，语气克制——预测未来用 may / could 而非 will，是社论保持分寸感的常规操作。
⑤ 反驳一：责任法救不了高速演进的 AI
这令人鼓舞，因为「让它冲」在实质层面同样站不住脚。特朗普的 AI 顾问们辩称，现行的责任法已经足够。他们说，对赔偿的恐惧、以及为赔偿投保的成本，让飞机制造商、汽车厂商和药企都规规矩矩。但面对 AI 高速演进的时代，法庭实在太迟缓了。在成熟行业里，官司都要打上好几年才有结果，更别提一个法官无先例可援引的全新行业。等法院作出判决时，AI 或许早已改变了整个经济——或者已经酿成一场大灾变。
本段反驳「现有法律够用了」这一论点。首句 That is encouraging, for “let it rip” fails on the substance, too 用 fails on the substance（在实质问题上站不住脚）承接上文（民意压力是好事），再转向「实质」。第二句 Mr Trump’s AI advisers argue that today’s liability laws are enough 用 argue that ... 引出对方主张——社论在反驳之前，必须先准确复述对方的观点，这样才显得有说服力。liability laws（责任法）是本段关键词。第三句 Fear of damages and the cost of insuring against them keep makers of aeroplanes, cars and drugs in line, they say 用 keep sb in line（使……守规矩）说明「赔偿压力」的约束作用，damages 是法律术语（损害赔偿金），insure against（为……投保）。第四句 But courts are too languid for the age of rapidly advancing AI 用 too ... for ...（对……来说太……）与 languid（迟缓无力）一语戳破「法律够用」的假象。第五句 Lawsuits take years to resolve in established industries, let alone a new one with no precedents for judges to call on 是本段最值得记的句型：let alone（更不用说）用于「前者都难，后者更不可能」的递进否定；precedents for judges to call on（法官可援引的先例）里 precedent（先例）是法律核心词，to call on（援引、利用）。末句 By the time they issue a ruling, AI might have transformed the economy—or caused a cataclysm 用 By the time ... 强调时间差，might have transformed（可能早已改造完毕）是对过去的推测，破折号后 or caused a cataclysm（或者已酿成大灾变）用比 catastrophe 更重的 cataclysm（大灾变）收束——「等判决下来，世界早已变样」，是「法律滞后论」最有力的一句话，值得整句背下。
⑥ 反驳二：「自我监管」为何站不住脚
特朗普「强力自我监管」的想法同样不是答案。「让它冲」那一派忘了：能够改变世界的新技术，总会招来新的规则。飞机需要检查与空中交通管制；汽车需要优先通行权、限速与安全带；新药必须经过测试，才能知道是否真的有效、是否安全。很难说清监管者究竟帮忙避免了多少起坠机、车祸或药物副作用。但几乎没有人会认为，这些领域里最优的监管量是零。
本段反驳「自我监管」论，用三个产业先例说明「新技术必得新规则」。首句 Mr Trump’s idea of “tremendous self-policing” is not the answer, either 里 self-policing（自我监管）是核心词，not ... either（也不）标明这是第二个被否定的方案。第二句 The “let it rip” crowd forget that world-changing new technologies always invite new rules 用 always invite new rules（总会招来新规则）给出全段论点，invite 此处意为「招致、引来」，比 need 更主动。接下来三句是全段最漂亮的排比：Aeroplanes need inspections and air-traffic control. Cars require right of way, speed limits and seat-belts. Novel drugs must be tested ...——飞机、汽车、新药三个领域，三种「必须」，用「名词短语堆叠」把「规则无处不在」讲得干净有力；注意作者刻意换了三个动词 need / require / must be 以避免单调，这是英语排比的高级做法。第五句 It is hard to say how many plane crashes, road accidents or medical side-effects regulators helped avert 用 how many 从句 + 省略 that/which 的定语从句（regulators helped avert），avert（避免、防止）是高分动词——真正的巧处在于：先承认「说不清救了多少」，反而更有力，因为后半句 But hardly anyone thinks the optimal amount of regulation in these areas is none at all. 用「几乎没人认为监管量应为零」轻巧收束。注意 hardly anyone（几乎没人）比 nobody 更有分寸，the optimal amount（最优数量）是经济学用语。
⑦ 警醒：还没出大事，但概率在上升
幸而，世界至今尚未发生一场「AI 切尔诺贝利」。在迄今曝光的攻击事件中最令人担忧的那桩 Hugging Face 攻击里，既没有人员伤亡，也没有真金白银的损失。但灾难的概率正在变大。再过几个月，开源模型的能力可能就赶得上 Anthropic 的 Mythos——首个真正擅长黑客攻击的模型。如果政客们非要等一场灾难降临才去写规则，那么灾后仓促起草的规则很可能很糟糕。眼下就已有大量坏主意：从「数据中心暂停令」到伯尼·桑德斯的「超级智能禁令」。而糟糕的法律也许只是开端：2001 年 9 月 11 日的恐怖袭击既促使人们反思航空安全，也带来了两场漫长的战争。
本段从「反驳对手」转向「警示风险」，是全文语气的转折点。首句 The world has, mercifully, so far avoided an AI Chernobyl 用 mercifully（幸而）作插入语，an AI Chernobyl（AI 界的切尔诺贝利）以一个人尽皆知的大灾难作比，一句话就让风险可感——用一个专有名词代替一段解释，是社论提高沟通效率的常用手段。第二句 No lives and no real money were lost in the Hugging Face attack, the most concerning to have come to light 用「没有死人、没有真金白银的损失」先诚实降低当前严重性，再用 the most concerning to have come to light（迄今曝光事件中最令人担忧的一起）把它重新拉回焦点。第三句 But the odds of disaster are shortening 用赌局用语 the odds ... are shortening（赔率在缩短，即越来越可能），比 getting higher 更形象。第四句 In a few months open models could be as capable as Anthropic’s Mythos, the first to be seriously adept at hacking 用 as capable as（能力可与……相当）+ adept at（擅长）给出具体时间表（几个月内），让「风险在逼近」有据可依。第五句 If politicians wait for a catastrophe before writing the rules, the ones they draft in haste in its aftermath are likely to be poor 是典型的「条件句 + 预测」：in haste（匆忙地）、in its aftermath（在灾后），the ones 代指 rules。第六句 Plenty of bad ideas are already out there, from data-centre moratoriums to Bernie Sanders’s “superintelligence ban” 用 from A to B 列举两端的坏主意，moratorium（暂停令）与 superintelligence ban（超级智能禁令）是本句两个具体例子。末句 And bad laws may be just the start: the terrorist attacks of September 11th 2001 inspired a rethink of aviation safety but also two long wars 用「9·11 既带来航空安全反思，也带来两场漫长战争」这一历史先例，警告「仓促立法可能付出更长期的代价」——用一个所有人都知道的历史事件做论据，比任何抽象论证都更有力。
⑧ 立论：好的监管该长什么样
明智的 AI 监管该是什么样子？目标绝不能是彻底阻挠技术进步。那样做的确会让民主的美国把技术前沿拱手让给威权主义的中国，并放弃非凡的科学与经济机遇。让 AI 实验室把自己的顶尖模型无限期地锁起来——即便这些模型能被控制住——也会在它们与其余所有人之间造成一道危险的鸿沟。
本段开始「给方案」，用一句设问与两组否定框出「好监管」的边界。首句 What might wise AI regulation look like? 用 What might ... look like?（……可能长什么样？）设问过渡——这是社论从「破」转向「立」的标准枢纽句。第二句 The goal cannot be to stymie technological advances altogether 先划出「不能做什么」：stymie（阻挠、妨碍）比 block 更书面，altogether（完全地）限定程度。第三句 That would indeed cause democratic America to cede the frontier to authoritarian China, and forgo remarkable scientific and economic opportunities 用 cause sb to do（使某人做）+ 两个并列动词 cede（让出）与 forgo（放弃）；cede the frontier to（把前沿阵地让给）是本句最重的短语——作者把「过严监管」的代价放进地缘政治的框架里说：不只是少赚点钱，而是把技术领导权让出去。democratic ... authoritarian ... 的对照进一步强化了这一分量。末句 Making AI labs lock up their cleverer models indefinitely ... could create a dangerous gap between them and everyone else 用动名词作主语（Making ... lock up ...），lock up（锁起来、封存）与 indefinitely（无限期地）描述另一种极端方案，a dangerous gap between them and everyone else（它们与其余所有人之间的危险鸿沟）点出该方案的副作用——本段示范了「立论」的一种高级写法：不急着说「该怎么办」，先把两种错误方案（放任不管 / 全部封存）都排除掉，剩下的空间就是正确答案。
⑨ 主张：测试、扣留、报告、保护、派驻
相反，美国应当强制要求：先进模型必须经过彻底测试，必要时还要扣住不放，直到它们拥有足以拦阻危险用途（比如制造生物武器）的保障措施为止。企业必须被要求上报安全事件，而代企业上报的吹哨人则应获得保护。独立评估员应当被派驻到各大实验室，就像银行监察员常驻放贷机构一样。
本段给出具体主张，是全文的「政策清单」。首句 Instead, America should mandate that advanced models be thoroughly tested and, if necessary, held back until they have safeguards to block dangerous use-cases, like building bio-weapons 是全篇最值得学的句式之一：mandate that + 主语 + (should) be done 用虚拟语气（mandate 等「命令/建议」类动词后的 that 从句用动词原形 be），这是雅思、考研写作的高分语法点；held back（扣住、不放行）与 have safeguards to block dangerous use-cases（具备拦阻危险用途的保障措施）构成「测试—扣留—有保障才放行」的三步。like building bio-weapons（比如制造生物武器）用一个具体例子解释 dangerous use-cases。第二句 Companies must be required to report safety incidents and whistleblowers who do so in their stead granted protection 用被动 must be required to（必须被要求）；in their stead（代替它们）里 stead 是古雅词，in sb’s stead = instead of sb；后半句 whistleblowers ... granted protection 省略了 must be（并列分句共享情态动词），即「吹哨人应获保护」。第三句 Independent evaluators should be embedded with major labs, just as bank examiners are with lenders 用 embed with（派驻到……内部）与 just as ... are with ...（正如……之于……）作类比——用「银行监察员常驻放贷机构」这一人人熟悉的制度，去解释「AI 评估员常驻实验室」这一全新提议，是让陌生政策瞬间可懂的经典操作。
⑩ 收束：把自愿变成强制，「Let it rip」是死路
这些想法中，有一些已在特朗普最近从各大 AI 公司那里劝来的自愿承诺里初露端倪。它们应当被明确写出，并变为强制要求。通过释放「认真监管 AI」的信号，美国还能促进国际合作，尤其是与中国的合作。模型在进步，规则也必须随之进步。「让它冲」是一条死路。
本段收束全文，从「国内立法」推到「国际合作」，最后以一句断言作结。首句 Some of these ideas were hinted at in the voluntary pledges that Mr Trump recently coaxed out of the big AI firms 用 be hinted at（被隐约提及）与 coax sth out of sb（从某人那里劝诱出某物）——coax（哄劝）用得极妙：暗示这些「自愿承诺」是特朗普连哄带劝才拿到的，含金量存疑。第二句 They should be spelled out and made mandatory 用 spell out（明确写出）+ made mandatory（变为强制）给出呼吁，mandatory 与首句 voluntary 形成「自愿 vs 强制」的对照。第三句 By signalling that it is serious about regulating AI, America would also foster international co-operation, particularly with China 用 By doing ..., S would ...（通过做……，就能……）这一「方式—结果」句式，signal 作动词「释放信号」、foster（促进、培育）都是高分词——这里有个耐人寻味的推理：只有自己认真管，别国才愿意跟你合作；「不管」反而会把自己孤立起来。末两句 As the models advance, so must the rules. “Let it rip” is a dead end. 是全文的「名言式收尾」：As A advances, so must B（A 前进，B 也必须前进）用 so must ... 倒装，把「规则必须跟上技术」压缩成一句对称的格言；最后一句 is a dead end（是一条死路）只四个词，用全篇最短的句子给整篇论证盖章——开头用 Let it rip 三字点题，结尾再用它三字宣判死刑，首尾同词、意义反转，是社论最漂亮的收束方式。
03核心词汇精讲
标签说明：四级 基础必备 六级 进阶 雅思 学术高频 高分 写作加分
A. 监管与政策：规则的语言
regulation/ˌreɡjuˈleɪʃn/四级雅思
释义n. 监管，管理；规章，条例
派生regulate v. 监管，管理；regulator n. 监管者；regulatory adj. 监管的
搭配AI regulation AI 监管；the optimal amount of regulation 最优的监管量；regulate an industry 监管一个行业
mandate/ˈmændeɪt/六级雅思
释义v. 强制要求，下令；n. 授权，命令
派生mandatory adj. 强制的；mandated adj. 被强制的
搭配America should mandate that ... 美国应当强制要求……；a clear mandate 明确的授权
moratorium/ˌmɒrəˈtɔːriəm/雅思高分
释义n. 暂停（期），暂缓（期）；正式禁令
派生moratorium（复数 moratoriums 或 moratoria）
搭配data-centre moratoriums 数据中心暂停令；a moratorium on new drilling 暂停新钻探
liability/ˌlaɪəˈbɪləti/六级雅思
释义n. 责任，法律责任；（复数）债务；累赘
派生liable adj. 有法律责任的；liability laws 责任法
搭配liability laws 责任法；accept liability 承担责任；the cost of insuring against damages 为赔偿投保的成本
safeguard/ˈseɪfɡɑːd/六级雅思
释义n. 保障措施，防护措施；v. 保护，捍卫
派生guard n./v. 守卫；safeguarding n. 保护
搭配have safeguards to block dangerous use-cases 具备拦阻危险用途的保障措施；safeguard privacy 保护隐私
pledge/pledʒ/四级雅思
释义n. 承诺，保证；v. 郑重承诺；抵押
派生pledge（过去式 pledged）；pledging n. 承诺
搭配voluntary pledges 自愿承诺；pledge to block some data centres 承诺拦下一些数据中心
mandatory/ˈmændətəri/六级雅思
释义adj. 强制的，强行的；法定的
派生mandate n./v. 命令，授权；mandatorily adv. 强制地
搭配made mandatory 变为强制；mandatory testing 强制测试；a mandatory rule 强制性规则
oversight/ˈəʊvəsaɪt/雅思高分
释义n. 监管，监督；（也指）疏忽，失察
派生oversee v. 监督；overseer n. 监督者
搭配independent oversight 独立监督；under the oversight of 在……的监督下
B. 风险与灾难：赌局的另一面
reckless/ˈrekləs/六级雅思
释义adj. 鲁莽的，不计后果的
派生recklessly adv. 鲁莽地；recklessness n. 鲁莽
搭配“let it rip” is reckless 「放手不管」是鲁莽的；a reckless gamble 一场不计后果的赌局
catastrophe/kəˈtæstrəfi/四级雅思
释义n. 大灾难，灾祸
派生catastrophic adj. 灾难性的；catastrophically adv. 灾难性地
搭配wait for a catastrophe 等待一场灾难；a catastrophic failure 灾难性的失败
cataclysm/ˈkætəklɪzəm/雅思高分
释义n. （尤指大范围的）大灾变，剧变
派生cataclysmic adj. 大灾变的，巨变的
搭配caused a cataclysm 酿成一场大灾变；a cataclysmic event 一场剧变
backlash/ˈbæklæʃ/雅思高分
释义n. （强烈的）反对，抵制，反弹
派生lash back 猛烈回击；counter-backlash n. 反向反弹
搭配the backlash against data centres 反数据中心的强烈反弹；a public backlash 公众的强烈反对
avert/əˈvɜːt/六级雅思
释义v. 避免，防止；转移（目光）
派生aversion n. 厌恶，回避；avertable adj. 可避免的
搭配regulators helped avert 监管者帮忙避免了；avert a disaster 避免一场灾难
cede/siːd/雅思高分
释义v. 割让，让出（领土、权利、地位）
派生cession n. 割让，让与
搭配cede the frontier to authoritarian China 把前沿让给威权主义的中国；cede ground 让出阵地
hazard/ˈhæzəd/四级雅思
释义n. 危险，危害；v. 使面临危险；冒险猜测
派生hazardous adj. 危险的；haphazard adj. 随意的
搭配a safety hazard 安全隐患；hazard a guess 冒险猜一下
onerous/ˈəʊnərəs/雅思高分
释义adj. 繁重的，费力的，麻烦的
派生onerousness n. 繁重
搭配overly onerous rules 过度繁重的规则；an onerous burden 沉重的负担
C. 政治与民意：选票的温度
bipartisan/ˌbaɪˈpɑːtɪzən/雅思高分
释义adj. 两党的，跨党派的
派生partisan adj. 党派的，偏袒的；partisanship n. 党派偏见；bipartisanship n. 两党合作
搭配bipartisan pressure 两党的压力；a bipartisan bill 一项两党共同提出的法案
midterm/ˌmɪdˈtɜːm/六级雅思
释义n. （美国）中期选举；adj. 期中的
派生midterm elections 中期选举；midterm exam 期中考试
搭配scrambled midterm races 搅乱中期选举的选情；midterm elections 中期选举
ploy/plɔɪ/雅思高分
释义n. 策略，花招，手段（常含贬义）
派生ploy（复数 ploys）
搭配a Chinese ploy to hobble America 中国用来束缚美国的招数；a marketing ploy 一种营销手段
toxic/ˈtɒksɪk/六级雅思
释义adj. 有毒的；（喻）极不受欢迎的，会惹麻烦的
派生toxicity n. 毒性；toxin n. 毒素；detox n. 排毒
搭配the “pro-AI” label is toxic 「亲 AI」标签有毒；a toxic political issue 一个碰不得的政治议题
lobby/ˈlɒbi/四级雅思
释义v. 游说（议员）；n. 游说团体；大厅
派生lobbyist n. 说客；lobbying n. 游说活动
搭配lobby the government 游说政府；tech lobbyists 科技行业说客
voter/ˈvəʊtə(r)/四级雅思
释义n. 选民，投票人
派生vote v./n. 投票；voting n. 投票；electorate n. 全体选民
搭配keep facing voters 继续面对选民；voters are growing concerned 选民越来越担心
administration/ədˌmɪnɪˈstreɪʃn/四级雅思
释义n. 政府，行政当局；管理，行政
派生administer v. 管理，执行；administrative adj. 行政的；administrator n. 管理者
搭配the Trump administration 特朗普政府；the attitude to AI in the administration 该政府对 AI 的态度
term-limited/ˈtɜːm ˈlɪmɪtɪd/雅思高分
释义adj. （因任期限制而）不能连任的，任期受限的
派生term n. 任期；term limit n. 任期限制；limit v./n. 限制
搭配the term-limited president 任期受限的总统；a term-limited governor 任期已满、不能连任的州长
D. 通用高分表达
ferociously/fəˈrəʊʃəsli/六级雅思
释义adv. 凶猛地，猛烈地
派生ferocious adj. 凶猛的；ferocity n. 凶猛
搭配advanced ferociously 以凶猛的势头推进；a ferocious debate 激烈的辩论
languid/ˈlæŋɡwɪd/雅思高分
释义adj. 迟缓的，无精打采的；慵懒的
派生languidly adv. 迟缓地；languor n. 倦怠
搭配courts are too languid for the age of AI 面对 AI 时代，法庭太迟缓了；a languid pace 慢吞吞的节奏
stymie/ˈstaɪmi/雅思高分
释义v. 阻挠，妨碍，使受挫
派生stymie（过去式 stymied）；stymieing n. 阻挠
搭配stymie technological advances 阻挠技术进步；stymied by red tape 被繁文缛节所阻
coax/kəʊks/雅思高分
释义v. 劝诱，哄劝；巧妙地取得
派生coaxing adj. 哄劝的；coaxingly adv. 哄劝地
搭配coaxed pledges out of the firms 从各公司劝来承诺；coax sb into doing sth 哄某人做某事
embed/ɪmˈbed/六级雅思
释义v. 嵌入，埋入；派驻（记者、专家等）
派生embedded adj. 嵌入的，内嵌的；embedment n. 嵌入
搭配evaluators embedded with major labs 派驻到各大实验室的评估员；an embedded journalist 随军记者
whistleblower/ˈwɪslbləʊə(r)/雅思高分
释义n. 吹哨人，举报人
派生blow the whistle on 举报；whistleblowing n. 举报行为
搭配whistleblowers granted protection 获得保护的吹哨人；a whistleblower complaint 举报人投诉
precedent/ˈpresɪdənt/六级雅思
释义n. 先例，前例；惯例
派生precede v. 先于，在……之前；precedence n. 优先（权）
搭配no precedents for judges to call on 法官无先例可援引；set a precedent 开创先例
adept/əˈdept/六级雅思
释义adj. 熟练的，擅长的；n. 内行，能手
派生adeptly adv. 熟练地；adeptness n. 熟练
搭配seriously adept at hacking 真正擅长黑客攻击；be adept at doing sth 擅长做某事
05写作积累
5.1 六个可直接套用的句型公式
5.2 话题语料（可直接搬进作文）
话题 1 · 科技监管与风险（Regulation & risk）
• 描述风险：the odds of disaster are shortening / the world has so far avoided an AI Chernobyl
• 反驳放任：It is hard to say how many ... regulators helped avert. But hardly anyone thinks the optimal amount of regulation is none at all.
• 法律滞后：courts are too languid for the age of rapidly advancing AI
• 提出主张：Instead, X should mandate that ...
• 国际合作：By signalling that it is serious ..., X would also foster international co-operation.
话题 2 · 公共政策与民意（Politics & public opinion）
• 民意转向：a quarter of Americans rated AI as an important issue. Now half do.
• 情绪合流：The backlash against data centres ... fuses tech-no-fear with NIMBYism.
• 政治后果：the “pro-AI” label is toxic.
• 预测未来：By next year, bipartisan pressure may build in Congress to do something.
• 复述对手：Advisers argue that ... / The “let it rip” crowd forget that ...
话题 3 · 论证与驳论（How to argue a case）
• 让步立论：X was flawed but defensible. But ... is reckless.
• 先例类比：Aeroplanes need ... Cars require ... Novel drugs must be ...
• 递进否定：..., let alone a new one with no precedents ...
• 风险警示：If politicians wait for a catastrophe before writing the rules ...
• 名言收尾：“Let it rip” is a dead end.
5.4 仿写练习：用「三例排比 + 递进否定」写一段「该不该管」的议论
【中文情境】每逢新技术出现，总有人说「先别管」。可这条路我们走过：铁路需要信号与时刻表，航空需要安检与塔台，药品需要临床试验。连成熟行业都难以约束的事，更难指望一个刚起步的行业靠自律。与其等一场灾难来倒逼立法，不如现在就定下最基本的规矩。
【英文示范】Every time a new technology arrives, someone says “leave it alone”. But we have been here before: railways need signals and timetables; aviation requires security checks and control towers; medicines must be trialled before they reach a single patient. What is hard to enforce in a mature industry is even harder to expect from a young one that polices itself. Rather than wait for a catastrophe to force legislation, we should set the basic rules now.
三处节奏：① 用「铁路 / 航空 / 药品」三个先例排比（公式 2），三个动词 need / require / must be 各不相同，避免单调；② 用 even harder to expect from 完成递进否定（公式 3 的变体）；③ 用 Rather than wait for ..., we should ...（与其等……不如……）收束，直接给出主张。结尾 force legislation（倒逼立法）是政策评论的高频搭配。
06篇章结构、论证手法与文化冷知识
6.1 论证结构（立论 → 让步 → 民意 → 政治 → 反驳 → 警醒 → 立论 → 主张 → 收束）
① 立论
特朗普政府对 AI 的态度停在 2025 年初——只有三个字：「Let it rip」
↓
② 让步
两年前情有可原（风险还在纸面），如今已属鲁莽——先给台阶再推翻
↓
③ 民意
四分之一 → 一半；数据中心遭抵制；Hugging Face 攻击与辞职的科学家
↓
④ 政治
连最铁杆的 MAGA 也发现「亲 AI」标签有毒，两党压力正在累积
↓
⑤ 反驳一
责任法够用？——法庭太慢，等判决下来世界早已变样
↓
⑥ 反驳二
自我监管？——飞机、汽车、新药三大先例：新技术必得新规则
↓
⑦ 警醒
尚未出大事，但概率在上升；灾后仓促立法代价更大（9·11 的双面教训）
↓
⑧ 立论
好的监管长什么样——先排除两种错误方案（放任 / 全封存）
↓
⑨ 主张
测试、扣留、报告、保护吹哨人、派驻评估员
↓
⑩ 收束
把自愿承诺变成强制规则，促进国际合作——「Let it rip」是一条死路
6.2 七种论证手法（可迁移到自己的议论文）
| 手法 | 原文例子 | 效果 |
|---|
| 让步再反转 | that approach was flawed but defensible. ... But in late 2026 “let it rip” is reckless | 先承认对方过去有理，再以时间点完成反转，驳论显得公允而有层次，不流于喊口号 |
| 先例类比 | Aeroplanes need ... Cars require ... Novel drugs must be tested ... | 用三个业已成熟的监管先例，把「AI 也该被管」变成不言自明的常识 |
| 数据与民意 | a quarter of Americans rated AI as an important issue. Now half do | 用两个数字的变化代替「越来越多人担心」这类空话，民意转向一目了然 |
| 专有名词作比喻 | the world has, mercifully, so far avoided an AI Chernobyl | 用一个众所周知的大灾难代替一段解释，风险立刻可感，沟通效率极高 |
| 俏皮祈使句 | Squint and ... looks like a trial run for a robot takeover | 用「你眯眼一看」引导读者自己得出结论，戏谑中藏锋芒，是《经济学人》的招牌笔法 |
| 递进否定 | Lawsuits take years to resolve in established industries, let alone a new one ... | 前句已足，后句更甚，把论证推向极端情形，反衬旧办法的无力 |
| 首尾同词反转 | 开头 “Let it rip.” 三字点题；结尾 “Let it rip” is a dead end. 三字宣判 | 同一短语两次出现、意义反转，收束干净有力，读者印象最深 |
6.3 文化冷知识
「MAGA」是什么
MAGA = Make America Great Again（让美国再次伟大），是特朗普的政治口号，也用来指代拥护他的那部分选民与政治势力。新闻里常把「运动/阵营」略作一个词来用，比如文中的 Who can blame MAGA ...（谁会怪 MAGA 呢）——用「阵营」代替一大群人，是英美政治报道的惯用简写。
「Let it rip」：三个字里的放任主义
Let it rip 原意是「松开手、让它尽情冲」（多用于发动机、汽车，即「一踩到底、别踩刹车」），引申为「不加约束、任其发展、放手一搏」。用在 AI 政策上，它指「先不监管，让技术自己跑」——与硅谷口号 move fast and break things（快速行动、打破旧规）一脉相承，是放任主义的通俗说法。
「AI 切尔诺贝利」：为什么用核事故做类比
切尔诺贝利（Chernobyl）是 1986 年苏联核电站事故，如今已成「技术失控」的代名词。称 AI 的潜在灾难为 an AI Chernobyl，等于在说：我们至今只是运气好。科技评论中类似的历史类比还有 AI 领域的 Three Mile Island（三里岛，1979 年美国核事故）与 Oppenheimer moment（奥本海默时刻，指科学家在灾难前夜的道德焦虑）。
NIMBYism：从「别建在我家后院」到数据中心
NIMBY = Not In My Back Yard（别建在我家后院），指「原则上支持，但别建在我这儿」的居民反对运动，汉语称「邻避主义」。文中说反数据中心运动 fuses tech-no-fear with NIMBYism，即「技术恐惧」与「邻避情绪」合流。数据中心耗电耗水、噪音不小，在许多地方已成为地方选举议题。
美国中期选举（midterm）：为什么共和党人开始怕
美国每两年改选众议院全部席位、参议院约三分之一席位；总统任期中途的这场选举称 midterm（中期选举）。总统任期受限（term-limited，连任两届后不能再竞选），所以议员比总统更在意民意风向。当「亲 AI」标签开始掉票，连得克萨斯州竞选参议员的肯·帕克斯顿（Ken Paxton）这样的铁杆 MAGA 人物，也承诺「拦下一部分数据中心」。
伯尼·桑德斯与「超级智能禁令」
伯尼·桑德斯（Bernie Sanders）是美国佛蒙特州参议员、民主社会主义者，曾两度角逐民主党总统提名。文中把他主张的「超级智能禁令」（superintelligence ban）与「数据中心暂停令」（data-centre moratoriums）并列为「坏主意」，意思是：激进的禁令同样会伤害创新——过于严苛与完全放任，两头都不可取。
「boffin」与英国媒体的小词
boffin 是英式俚语，指科学家、技术怪才（略带戏谑），文中用它形容那位从 Anthropic 辞职、指责公司「拿我们的命在赌」的研究者。同类小词还有 egghead（书呆子、知识分子）、wonk（政策专家、技术官僚）等，都是英国媒体偏爱、能让严肃报道显得轻松的词汇。