根据《金融时报》的分析，Anthropic 今年在高级 AI 的危险性方面发出警告的频率远高于竞争对手 OpenAI，而批评者指责该公司助长了美国禁止外国用户访问其最新模型的行为。
《金融时报》的研究分析了 Anthropic 或其首席执行官达里奥·阿莫迪撰写的官方声明、社交媒体帖子和文章，发现该公司在 2026 年使用的每 1000 个单词中，有 5 个与风险、监管或限制相关。OpenAI 和山姆·奥尔特曼的相应数字则低八倍，为每 1000 个单词中出现 0.6 个。
在华盛顿上周禁止外国公民使用 Anthropic 的最新模型 Mythos 和 Fable 之后，这种比较变得具有政治色彩。一些技术专家将这一决定归咎于这家估值 9650 亿美元的 AI 集团反复就 AI 对社会造成的风险发出警告——尤其是与 Mythos 相关的风险。
Meta 前首席 AI 科学家、AI 先驱之一扬·勒昆本周表示，出口禁令表明阿莫迪关于 AI 的“荒谬的恐吓宣传”终于得逞了。“种瓜得瓜，种豆得豆，”他在一周前的一篇社交媒体帖子中写道。
这场争议已引起欧洲和硅谷部分地区的警惕，那里的高管和官员担心特朗普政府可能愿意限制非美国用户访问前沿模型。这正成为考验美国打算如何监管日益强大的 AI 模型的早期测试。
《金融时报》创建了包括“有害”、“危险”和“未对齐”在内的术语列表，并计算了这些术语在每家公司或其 CEO 的声明中出现的频率。它还使用情感分析来比较沟通内容的积极和消极语气。
研究发现，在 Anthropic 2026 年的沟通内容中，“风险”出现了 336 次；“保障措施”出现了 121 次；“漏洞”出现了 128 次。在 OpenAI，这些术语分别被使用了 30 次、33 次和 10 次。
在四月份的一次播客采访中被问及 Mythos 时，奥尔特曼说：“说‘我们造了一颗炸弹。我们正准备把它扔到你头上’，这显然是绝妙的营销手段。”
Anthropic 一直试图将自己定位为 AI 行业的良心。它频繁发布关于该技术潜在危害的研究论文和声明，并多次呼吁政府加强干预。
在出口禁令颁布前几天，Amodei 在其个人网站上发布了一篇长篇博文，指出监管机构在 AI 方面的行动过于迟缓。
“在过去几个月里……AI 惊人能力及其风险的证据已变得不容否认，”他写道，并补充说 Mythos 展示了“对网络安全非常真实的威胁，有可能扰乱金融部门、关键基础设施和国家安全。”
但这篇帖子远非《金融时报》所分析的 Anthropic 最“末日论”的出版物。一篇关于 2025 年 4 月 AI 罕见行为（如破坏或提供武器相关信息）的新闻稿，其负面措辞大约是其三倍。
《金融时报》的分析还发现，自 2023 年以来，随着其 AI 工具日益普及，Anthropic 的措辞已显著软化。其使用与风险和监管相关语言的频率，相比 2023 年同期大约减少了一半。
总体而言，数据显示，虽然 Anthropic 的公开沟通基调总体上是积极的，但其积极程度不及 OpenAI。
-
1. 评测：《寡妇湾》是对喜剧恐怖片的一次大胆原创尝试 - 2. 火箭报告：蓝色起源发射台重建工作启动；相对论公司瞄准火星
- 3. 伯尼·桑德斯公布 7 万亿美元计划，旨在让美国民众掌控 AI 行业
- 4. FDA 顾问委员会在机构风波后一致投票批准 Moderna 的 mRNA 疫苗
- 5. 特朗普政府的煤炭投资为屡次违规的工厂提供支持
Anthropic has warned about the dangers of advanced AI far more often than rival OpenAI this year, according to FT analysis, as critics accuse the company of helping to trigger a US ban on foreign access to its newest models.
Five in every 1,000 words used by Anthropic in 2026 related to risk, regulation, or restrictions, according to FT research that analyzed official statements, social media posts, and articles written by the company or its chief, Dario Amodei. The equivalent figure for OpenAI and Sam Altman was eight times lower, at 0.6 words per 1,000.
The comparison has become politically charged after Washington last week barred foreign nationals from using Anthropic’s latest models, Mythos and Fable. Some technologists have blamed the decision on the $965 billion AI group’s repeated warnings about AI’s risk to society—particularly in relation to Mythos.
Yann LeCun, Meta’s former chief AI scientist and one of AI’s pioneers, said this week the export ban showed that Amodei’s “ridiculous fear-mongering” about AI had finally paid off. “One reaps what one sows,” he wrote in a social media post a week ago.
The dispute has alarmed parts of Europe and Silicon Valley, where executives and officials fear the Trump administration may be willing to restrict non-US access to frontier models. It is emerging as an early test of how the US intends to oversee increasingly powerful AI models.
The FT created lists of terms including “harmful,” “dangerous,” and “misaligned” and calculated how frequently they appeared in statements by each company or its CEO. It also used sentiment analysis to compare the positive and negative tone of communications.
The research found that in Anthropic’s communications in 2026, “risk” appeared 336 times; “safeguard,” 121 times; and “vulnerability,” 128 times. At OpenAI, these terms were used 30, 33, and 10 times, respectively.
When asked about Mythos in a podcast interview in April, Altman said: “It is clearly incredible marketing to say, ‘We have built a bomb. We are about to drop it on your head.’”
Anthropic has long sought to position itself as the AI industry’s conscience. It frequently releases research papers and statements about the potential harms of the technology. It has also repeatedly called for greater government intervention.
Days before the export ban, Amodei published a long blog post on his personal website arguing that regulators were moving too slowly on AI.
“In the last few months… the evidence of AI’s incredible power, as well as its risks, has become undeniable,” he wrote, adding that Mythos demonstrated “very real risks to cyber security, creating the potential for disruption of the financial sector, critical infrastructure, and national security.”
But the post was far from the “doomiest” Anthropic publication analyzed by the FT. A news item from April 2025 about rare AI behaviors, such as sabotage or providing information about weapons, contained roughly three times more negative language.
The FT analysis also found that Anthropic had softened its language significantly since 2023, as its AI tools had gained popularity. Its use of risk and regulation-related language has roughly halved from the same period in 2023.
Overall, the data showed that while Anthropic’s public communications were largely positive in tone, they were less so than OpenAI’s.