Anthropic公司宣布将在Claude中嵌入隐形水印标记，以识别AI生成的文本1。这一举措应对了学生日益广泛使用AI进行学术不诚实行为的问题。根据2026年英国调查，95%的本科生使用AI，其中94%在评估中使用1。澳大利亚大学的情况同样令人担忧，Turnitin在2024年7月的报告显示，超过53%的提交作业使用某种形式的AI1。
然而，水印技术存在根本性局限。该技术无法检测AI是否仅被用于校对、编辑改进，也无法识别来自不同AI系统（如DeepSeek）的文本1。OpenAI因担心对非英语使用者等群体产生负面影响，未公开推出其文本水印工具1。与此同时，欧盟新AI法案规定AI系统提供商必须标识合成文本、音频、视频或图像1。
面对水印技术的不足，教育工作者提出了不同的应对思路。昆士兰大学教育心理学专家Jason Lodge建议停止寻找学生使用AI作弊的证据，转而寻找学习成果的证据1。澳大利亚高等教育监管机构也建议采用连贯递进的评估方法，包含至少一项"安全"评估任务的教学方案1，以此来验证学生真实的学习能力，而非仅依赖技术手段进行检测和惩罚。
Anthropic has announced plans to embed invisible watermarks in Claude outputs to identify AI-generated text, marking an attempt to address widespread concerns about students using artificial intelligence to cheat on assignments 1. The European Union's new AI legislation requires AI system providers to label synthetic text, audio, video, and images, reflecting regulatory pressure to make AI generation more transparent 1. However, experts warn that watermarking technology faces significant limitations and may not effectively solve academic integrity challenges.
The scale of AI use in academic settings has grown substantially. A 2026 UK survey found that 95 percent of undergraduate students use AI, with 94 percent employing it during assessments 1. In Australian universities, over 53 percent of submitted assignments contained some form of AI usage, according to a Turnitin report from July 2024 1. Despite these figures, watermarking cannot detect whether AI was used for proofreading or editorial improvement, nor can it identify which AI system generated the text—such as distinguishing between Claude and DeepSeek 1. OpenAI has declined to publicly release its own text watermarking tool due to concerns about negative impacts on non-English speakers and other populations 1.
Rather than pursuing detection and punishment strategies, educators are pivoting toward alternative approaches. Jason Lodge, an educational psychology expert at the University of Queensland, recommends that institutions stop seeking evidence of student AI cheating and instead focus on verifying learning outcomes 1. Australian higher education regulators have suggested implementing coherent, progressive assessment methods that include at least one "secure" evaluation task, moving away from reliance on technology-based detection alone 1.
评论
还没有评论，欢迎留下第一条。