Why AI Companies Want You to Be Afraid of Them

You may have heard a story like this before. A technology company announces a powerful new AI. It claims the system is so capablecapable: Having the ability or qualities necessary to do something — 有能力的；能干的 that releasing it would be dangerous, so for now the company is keeping it lockedlock: To become fastened in place — 卡住 away. This is exactly what Anthropic is saying about its latest model, Claude Mythos. According to the company, Mythos can find security flawsflaw: A defect, weakness, or imperfection in something or someone. — 缺陷；瑕疵；弱点 in computer code better than human experts. The results could be severesevere: Very great in degree; strict or harsh. — 严重的；严厉的 for the economy and for national security if similar tools end up in the wrong handsin the wrong hands: in the possession of people who would use it for harm — 落入恶人手中.

Some security experts doubt these claims, but the bigger question is why AI executivesexecutive: A senior manager in a business or organisation; the branch of government responsible for implementing laws. — 高管；行政人员 so often warn about their own products. It is a strange way for any businessbusiness: A specific commercial enterprise or establishment — 企业;商号 to behave. You do not hear McDonald's announce that it has created a burger so dangerously tasty that selling it would be wrong. Yet leaders at the largest AI firmsfirm: A business or company, especially a small one. — 公司；商行 keep telling the public that their tools may destroy society. If they trulytruly: In a truthful way; genuinely; to the fullest degree. — 真正地；确实地 believe this, why are they still building and selling them?

Criticscritic: A person who expresses an unfavourable opinion; a professional reviewer of art, literature, or music. — 批评家；评论家 offer one answer. Shannon Vallor, a professor of AI ethicsethic: A set of moral principles governing behaviour; moral philosophy. — 伦理；道德规范 at the University of Edinburgh, argues that talking about an AI apocalypseapocalypse: A disaster; a cataclysmic event — 灾难；灾难性事件 keeps customers and lawmakerslawmaker: One who makes or enacts laws — 立法者；议员 focused on a distantdistant: Far away in space or time; not closely related; reserved or unfriendly. — 遥远的；疏远的 threatthreat: A statement of intention to inflict harm; a person or thing likely to cause danger — 威胁；恐吓. Meanwhile, the harmharm: Physical or mental damage or injury. — 伤害；损害 these systems are already causing goes ignored. This pattern, she suggests, is a form of fear mongeringfear mongering: deliberately spreading fear, often to influence people for one's own advantage — 贩卖恐惧: companies exaggerateexaggerate: To represent something as being greater or more extreme than it actually is. — 夸大；夸张；言过其实 the power of their products to lift stockstock: A supply of goods kept for sale; shares in a company; an amount available for use. — 库存；股票；储备 prices and to send a useful message to governments. The message is that only the AI firms themselves can be trustedtrust: To place confidence in; to rely on, to confide, or have faith, in — 信任；信赖 to manage the danger, so regulatorsregulator: A person or group that sets standards of practice, especially those established by law — 监管机构 should step back. Sam Altman, the head of OpenAI, recently called Anthropic's approachapproach: A way of dealing with a situation or problem; a method — 方法；途径 fear-based marketingfear-based marketing: promoting a product by emphasising danger or risk to the audience — 恐惧式营销, although Altman has used very similar language about his own products for years.

Emily Bender, a linguisticslinguistics: The scientific study of language — 语言学 professor at the University of Washington and co-author of the book The AI Con, sees the same pattern. She says the constantconstant: Occurring continuously over a period of time; unchanging — 持续的；不变的 talk of extinctionextinction: The action of making or becoming extinct; annihilation — 灭绝；消亡 directs attention away from problems that already exist, such as environmental damage, exploitedexploit: To use a situation or resource to gain an unfair advantage; to utilise fully. — 利用；剥削 workers, and social systems being weakenedweaken: to make or become less strong, powerful, or effective — 削弱；减弱；使变弱 by these tools. Other researchers question whether Anthropic's safety claims about Mythos can even be checked. Heidy Khlaaf of the AI Now Institute notes that Anthropic has not shared standard measurementsmeasurement: The act of measuring; a quantity, dimension, or capacity ascertained by measuring. — 测量；尺寸 security engineers normally relyrely: To depend on someone or something with full trust or confidence. — 依赖；依靠 on. The threat may be real, she says, but the evidence so far is thin.

The history of these companies makes the warnings harder to trust. OpenAI began as a non-profit, meaning it was set up to serve the public rather than make money for owners. Later, a splintersplinter: A group that formed by splitting off from a larger membership — （脱离的）分裂派；分裂出的小团体 groupsplinter group: a small group that has broken away from a larger organisation due to disagreement — 分裂出去的小派系 left OpenAI and foundedfounded: to start an institution or organization — 创立；建立（机构或组织） Anthropic, saying that their old employer was not careful enough about safety. Today both are working to become publicly traded companies, which would let investorsinvestor: A person or organisation that puts money into financial schemes or businesses with the expectation of profit. — 投资者 buy sharesshare: A financial instrument that shows that one owns a part of a company that provides the benefit of limited liability — 股票;股份 in them on the stock market. At the same time, Google has dropped its red linesred lines: limits or rules a person or organisation has promised not to cross — 底线 / 红线 against helping to build AI weapons, and Anthropic has quietly abandonedabandon: To leave a place, thing, or person permanently; to give up completely — 放弃；遗弃 its old promise never to train a model without strong safety guarantees. As Vallor puts it, you should not expect any of these firms to walk away from market dominancedominance: Power and influence over others; the state of being dominant or having control. — 主导地位；支配；优势 just to remain the good guys.

The same executives who warn of disaster also promise paradiseparadise: A very pleasant place; a place full of lush vegetation — 乐土；人间天堂: a curecure: To relieve a person or animal of the symptoms of a disease; to solve a problem. — 治愈；治疗 for the climate crisiscrisis: A time of intense difficulty or danger; a turning point. — 危机, coloniescolony: A country or area under the political control of another country; a group of organisms living together. — 殖民地；群体 in space, a future of brilliant ideas. Vallor argues that apocalypse and utopiautopia: A world in which everything and everyone works in perfect harmony — 一个一切事物和每个人都完美和谐运作的世界 are two sides of the same coin. Both make AI feel too vastvast: Of very great extent or quantity; immense. — 广阔的；大量的 for ordinary rules to touch, which suits the companies very well. Yet AI tools are not gods. They are products built for profit, and societies have regulatedregulate: To control or supervise something by means of rules and regulations. — 管理；调节；规范 things far more powerful, from nuclear weapons to biologicalbiological: Relating to biology or living organisms. — 生物的；生物学的 research. Deepfakesdeepfake: A photo or video created using artificial intelligence, that imposes the face of one person onto the body of another — 深度伪造（影像） may have crossed the point of no returnpoint of no return: a stage past which it is impossible to undo or reverse a situation — 无法回头的临界点, but most AI harms have not. We were once told that Bitcoin would replace world currencies and that social media would save democracydemocracy: A system of government in which citizens exercise power through elected representatives — 民主；民主制度. Perhaps the latest predictions will come true. Or perhaps they will quietly fadefade: To gradually become weaker, less distinct, or less prominent over time. — 褪色；消退；逐渐消失.