最新数据显示，前Anthropic研究员雅各布·科克森帮助推入主流的AI安全叙事，数周后仍在塑造相关讨论。
New data shows the AI-safety narrative former Anthropic researcher Jacob Coxon helped mainstream is still shaping the conversation weeks later.
为何值得关注：一旦某种强大的框架站稳脚跟，后续的声音和事件即便不来自最初的信息发布者，也可能进一步强化它。
Why it matters: Once a powerful frame takes hold, subsequent voices and events can reinforce it even without the original messenger.
数据说明：根据PeakMetrics的数据，科克森辞职三周后，每日有关AI风险的讨论量仍保持在病毒式传播前水平的4.4倍。
By the numbers: Three weeks after Coxon's resignation, daily AI-risk conversation was still running at 4.4 times its pre-viral level, according to PeakMetrics.
然而，科克森及其他研究人员的警告在相关讨论中的占比，已从9月9日至10日的62.5%降至9月21日至28日的13.4%。
Yet Coxon and other researchers' warnings have fallen from 62.5% of the conversation on Sept. 9–10 to just 13.4% by Sept. 21–28.
与此同时，有关政策、政治和市场的讨论占比从27.4%上升至61.5%。
Instead, discussion of policy, politics and markets rose from 27.4% to 61.5%.
整体来看：科克森时刻将一个颇具吸引力的个人故事与一个强大的框架结合起来：就连正在开发先进AI的人，也担心它将走向何方。
The big picture: The Coxon moment combined an appealing personal story with a powerful frame:Even the people building advanced AI are worried about where it's going.
Anthropic的其他研究人员公开强化了这一警告；新闻机构放大并审视这一警告；议员们则在简报和行动呼吁中采纳了这一观点。
Other Anthropic researchers publicly reinforced the warning; news organizations amplified and interrogated it; and lawmakers picked it up in briefings and calls for action.
各AI公司的首席执行官也呼吁保持克制，推动这一议题进入新一轮报道，使该框架持续受到关注。
AI CEOs responded with their own calls for restraint, sending the issue into another round of coverage and keeping the frame relevant.
典型例子是：Axios记者麦迪逊·米尔斯本周末独家披露，OpenAI、Anthropic及安全研究人员正在调查数万起涉及前沿模型的事件，这些模型的行为被评测人员认为存在问题。
Case in point: Axios' Madison Millsscoopedthis weekend that OpenAI, Anthropic and security researchers are investigating tens of thousands of incidents involving frontier models behaving in ways evaluators considered problematic.
英伟达首席执行官黄仁勋一直反对这种末日论式的框架，认为AI风险被夸大了，技术层面的安全措施才是解决方案。但黄仁勋仍在参与这场由科克森帮助界定的辩论。
Nvidia CEO Jensen Huang has been pushing back against the doomsday framing, arguing that AI risks are being overstated and that technological safeguards are the answer. But Huang is still engaging in the debate Coxon helped define.
西蒙的思考时刻：这让我想起我以前在乔治·华盛顿大学的教授罗伯特·恩特曼提出的一种传播理论框架。
Simon's thought bubble: This reminded me of a comms theory framework developed by my former George Washington University professor Robert Entman.
他认为，最具内在力量的框架，是那些与社会对某一议题的既有感受“相一致”的框架，以至于它们看起来就像常识。
The most inherently powerful frames, he argued, are those "congruent" with how society already feels about an issue, so much that it seems like common sense.
考克森设定的叙事框架之所以持续占据主导，是因为“未来主义者失去人工智能控制权”这一概念在《终结者》和《黑客帝国》等系列电影中反复出现。
Coxon's framing continues to dominate because the idea of futurists losing control of AI is commonly revisited in popular fiction, including "The Terminator" and "The Matrix" movie series.
具体而言：PeakMetrics 分析了8月25日至9月28日期间新闻报道和社交媒体中有关人工智能风险、人工智能安全、失控的人工智能代理或雅各布·考克森的内容，覆盖新闻报道、X、Reddit、TikTok、Instagram、Threads、Facebook和Bluesky。
The fine print: PeakMetrics analyzed news and social mentions of AI risk, AI safety, rogue AI agents or Jacob Coxon from Aug. 25–Sept. 28 across news coverage, X, Reddit, TikTok, Instagram, Threads, Facebook and Bluesky.
研究比较了8月25日至9月8日、9月9日至10日以及9月21日至28日三个时段，并利用人工智能按照叙事类型、作者立场和责任归因对相关提及进行分类。
It compared Aug. 25–Sept. 8, Sept. 9–10 and Sept. 21–28, using AI to classify mentions by narrative, author stance and blame.
归根结底：考克森和人工智能安全倡导者不仅多次获得顶级媒体曝光，还在传播层面赢得了主动权：我们仍在按照他们设定的框架讨论这个问题。
The bottom line: Coxon and AI safety advocates scored a comms win beyond multiple top-tier media placements: We're still debating the issue the way they framed it.