威斯敏斯特的政治政策写作正越来越多地展现出生成式人工智能的特征，例如依赖对比、排比（三部曲）以及过度使用破折号。这些独特的文体模式现在很容易被复杂的AI检测工具（如Pangram）识别，根据芝加哥大学研究人员的说法，该工具声称其误报率仅为0.01%，且在较长文本上的错误率接近于零。使用Pangram，调查人员标记了多份政策文件，包括国会议员路易斯·海格（Louise Haigh）的报告，以及工党团体“主流”（Mainstream）发表的一份长达70页、名为《生产性国家》的宣言，这些文件被判定为主要由AI生成。作者们为其使用AI进行了辩护，声称这仅用于版本控制，或在使用Claude进行文案编辑时统一不同草稿的语气。

在起草政治政策时，AI的使用日益盛行，这引发了人们对政治思想和交流质量的严重担忧。虽然政客在历史上一直依赖演讲撰稿人或研究助理，但将写作外包给像Claude或ChatGPT这样的AI模型，存在绕过将思想付诸纸面的关键（尽管有时是痛苦的）过程的风险，而写作过程自然会精简并加强论点。通过让AI处理写作，作者面临着写出掩盖混乱思维的平庸散文的风险。此外，由于读者无法判断AI是仅用于最后的润色还是用于生成核心思想，在被媒体质问之前不公开AI的参与会损害公众的信任。

在试图影响候任首相安迪·伯纳姆（Andy Burnham）的努力中，这种技术捷径尤为明显，因为伯纳姆在重大问题上依然含糊其辞。由于“伯纳姆主义”仍然定义模糊，政治创业者们正急于通过威斯敏斯特传统的长篇PDF宣言来塑造他的议程。由于这些文件必须迅速产出以抓住政治机遇，作者们越来越多地求助于Claude来生成大量的政策文章。这一趋势表明，虽然AI可以加快政治观点的产出，但它可能会同时降低国家治理框架的知识深度。

Political policy writing in Westminster is increasingly displaying the hallmarks of generative AI, such as a reliance on contrasts, tricolons, and excessive em-dashes. These distinct stylistic patterns are now easily identified by sophisticated AI detection tools like Pangram, which boasts a remarkably low 0.01% false-positive rate and a near-zero error rate on longer texts according to researchers at the University of Chicago. Using Pangram, investigators flagged multiple policy documents, including papers by MP Louise Haigh and a 70-page manifesto titled "The Productive State" by the Labour group Mainstream, as being largely AI-generated. The authors defended their use of AI by claiming it was used for version control or to copy-edit and unify separate drafts using Claude.

The rising prevalence of AI in drafting political policy raises critical concerns about the quality of political thought and communication. While politicians have historically relied on speechwriters or research assistants, outsourcing writing to AI models like Claude or ChatGPT risks bypassing the crucial, albeit difficult, process of putting thoughts onto paper, which naturally refines and tightens arguments. By letting AI handle the writing, authors risk producing clunky prose that papers over sloppy thinking. Furthermore, because readers cannot determine whether AI was used merely for a final polish or for generating core ideas, failing to disclose AI involvement until confronted by the media damages public trust.

This technological shortcut is particularly visible in efforts to influence Andy Burnham, Britain's prime-minister-in-waiting, who remains vague on major issues. Because Burnhamism is still loosely defined, policy entrepreneurs are rushing to shape his agenda through the traditional Westminster currency of lengthy PDF manifestos. Since these documents must be produced rapidly to seize political opportunities, authors are increasingly turning to Claude to generate high volumes of policy prose. This trend suggests that while AI may speed up the output of political ideas, it may simultaneously degrade the intellectual depth of the country's governing frameworks.

Source: Is AI writing taking over Westminster?

Subtitle: Several influential policy papers are failing AI detectors

Dateline: Jul 09, 2026 05:22 AM