雅思阅读 62: The Forged Voice（被伪造的声音）

改编自 European Commission / 中国最高人民法院 / 各国议会文献（2025-2026）。雅思阅读 Section 3 难度，约 1050 词。 素材来源：https://english.news.cn/20260907/acf67ea35acf4eb989d1a6701d696207/c.html

Reading Passage

A. A deepfake is, at its core, an ordinary lie told by extraordinarily cheap means. Using machine-learning models trained on thousands of hours of a person's face, voice and mannerisms, almost anyone can now generate a video in which that person appears to say or do something they never said or did, and distribute it to a global audience within minutes. Only a few years ago such technology required specialised equipment and a team of technicians; today it is available through apps and open-source toolkits on an ordinary laptop. The phrase itself combines "deep learning" with "fake," and it has moved, in under a decade, from a novelty used by filmmakers and pranksters to a tool of extortion, defamation and political interference. The reason legislators are now scrambling is not that the technology has suddenly become more dangerous, but that it has become too ordinary for existing law, written for the age of the edited photograph, to handle. A photograph can be faked, but for most of the twentieth century faking it well required real skill; a convincing video now requires only a dataset and an evening.

B. Elections were the first target. In Texas, in 2019, the state legislature passed Senate Bill 751, which made it a criminal offence to create or distribute a deceptive synthetic video within thirty days of an election if the intention is to injure a candidate or influence the result. The law's definition is telling: a deepfake is treated as something created with the deliberate intent to deceive about a real person's conduct. Other jurisdictions have moved more recently. In March 2026, the government of New South Wales, Australia, introduced new electoral offences specifically covering AI-generated material, ahead of the state's 2027 election, arguing that synthetic media could be used to manipulate opinion or unfairly damage a candidate. In France, the SREN law now treats the sharing of deepfakes on social media as an aggravating circumstance, punishable by up to two years' imprisonment and a fine of 45,000 euros. In the United States, the AI Transparency in Elections Act of 2026 would require political advertisements that contain AI-generated images, audio or video to say so explicitly. The common thread is timing: lawmakers treat the weeks before a vote as uniquely fragile, and try to ring-fence them.

C. Beyond elections, the harder problem is the impersonation of ordinary people. In September 2026, China's Supreme People's Court issued guidelines addressing a wave of disputes that had quietly arrived in the courts: AI deepfakes, voice cloning, algorithmic price discrimination and AI-generated false information. The guidelines were unambiguous: no one may use AI to create or distribute a recognisable digital replica of another person — a cloned face, a synthetic voice — without that person's consent. Doing so, the court said, infringes the individual's personality rights, and using the replica to spread false claims or damage a reputation can bring civil liability on top. The ruling matters because it names a harm that earlier laws did not quite reach. A deepfake of a private citizen, used to harass them or to sell a product in their likeness, does not have to alter an election to cause real damage. It does not even have to be believed. The fact that a video of you, apparently, exists saying things you never said is itself a burden that the law is now reluctantly being asked to lift.

D. Europe has approached the issue through a single, wide-ranging instrument rather than a patchwork of election laws. The EU's Artificial Intelligence Act, which entered into force in stages, has prohibited deliberately deceptive manipulative AI systems since February 2025, and from August 2026 it imposes transparency obligations on the rest. Under those rules, interactive chatbots must tell users that they are talking to a machine; AI-generated or heavily edited images, video and audio must be labelled as synthetic; and the content must carry a machine-readable mark so that platforms, not only human viewers, can recognise it. The European Commission argued in mid-2026 that these provisions would reduce deception and manipulation without banning the technology itself. The model is in one sense liberal: synthetic media is permitted, but it must identify itself. In another sense it is a bet — that labels and watermarks will keep pace with generators that are improving every month, an arms race that detection tools have not yet convincingly won.

E. That arms race is the part of the problem that no law has yet solved. Detection software is forever playing catch-up: a detector trained on this year's generators is blind to next year's. Platforms can label content retroactively, but they cannot un-make a video that has already been downloaded, remixed and reposted a million times. And even when the law is clear — you may not clone a person's face or voice without consent — proving who made a deepfake, and where, across jurisdictions and pseudonymous accounts, remains fiendishly difficult. The most likely outcome is therefore not a world without synthetic media but one in which the burden of proof shifts: where a deepfake of a politician, an actor or an ordinary citizen is treated less as a neutral piece of content and more as a forged document — something that may be fascinating, even artistic, but that no one is obliged to take at face value. Forgery, after all, did not disappear when paper money was invented. It merely learned to live alongside the institutions designed to detect it. The same will be true of the synthetic image: not banned, not trusted, but treated as a document that must be authenticated, sourced and taken seriously only when its provenance can be verified. For the first time, ordinary citizens — not just archivists and journalists — will have to learn the reflex that a video is no longer, by default, evidence.

Questions 1-4

Choose the correct heading for paragraphs B, C, D and E from the list of headings below.

List of Headings i. The election-cycle response to synthetic media ii. Protecting ordinary people's likeness and voice iii. The European transparency model iv. Why detection is an arms race, not a finish line v. A brief history of film special effects vi. How smartphones are manufactured vii. The economics of online advertising

1. Paragraph B: ____
2. Paragraph C: ____
3. Paragraph D: ____
4. Paragraph E: ____

Questions 5-8

Choose the correct letter, A, B, C or D.

5. From when do the EU AI Act's transparency obligations for deepfakes apply? A. August 2026. B. February 2019. C. January 2020. D. They never apply.

6. What did Texas's Senate Bill 751 (2019) make a criminal offence? A. Creating or distributing a deceptive synthetic video within 30 days of an election to influence it. B. Taking any photograph of a politician. C. Owning a smartphone. D. Posting any message on social media.

7. What did China's Supreme People's Court guidelines (September 2026) state? A. People may not create recognisable digital replicas of others without consent. B. Deepfakes are freely permitted in all contexts. C. All AI research must immediately cease. D. Only the government may use AI.

8. Under France's SREN law, what can sharing a deepfake on social media lead to? A. Up to two years' imprisonment and a €45,000 fine. B. A small tax per view. C. A mandatory apology on television. D. No penalty at all.

Questions 9-13

Do the following statements agree with the claims of the writer?

Write:

• TRUE if the statement agrees with the information
• FALSE if the statement contradicts the information
• NOT GIVEN if there is no information on this

9. Since February 2025, the EU AI Act has prohibited deliberately manipulative, deceptive AI systems.
10. Under the new EU rules, chatbots must inform users that they are interacting with AI.
11. Australia has passed no laws at all concerning political deepfakes.
12. The U.S. AI Transparency in Elections Act of 2026 has already been signed into law.
13. Most deepfake videos are used exclusively to produce Hollywood feature films.

Questions 14-15

Complete the summary below using NO MORE THAN TWO WORDS from the passage.

The EU requires that AI-generated images, video and audio be (14) __________ as synthetic, while Chinese courts ruled that creating recognisable digital (15) __________ of another person without consent infringes their rights.

答案与解析

← 上一篇 | 返回雅思焦点 | 下一篇 →