科技评论家Om Malik撰文批评Anthropic公司将其最强大AI模型命名为"Mythos"的举动，认为这一选择反映了该公司采用"神话叙事"而非"理性论证"的策略[1]。根据Malik的分析，"Mythos"这个源自古希腊的词汇意为"接收而非验证的故事"[1]。Malik指出，Anthropic通过White House访问、Vatican拜访等精心设计的公关活动，为其AI技术建立道德合法性，同时规避透明度和外部审查[1]。
Anthropic声称Mythos因网络安全原因不向公众开放，仅通过"Project Glasswing"向少数受信任机构提供[1]。Malik指控该公司通过道德表演和受控披露来积累信誉，进而规避真正的外部验证[1]。他将Anthropic的权力运作与Augustus的统治、中世纪君主专制的神圣王权以及"天定命运"进行了历史类比[1]。
Anthropic联合创始人Dario Amodei曾在CBS的《60 Minutes》节目中表达过对决策集中的担忧。他指出："我对这些决定由少数几家公司、少数人做出感到非常不安。没有人选举我。"[1]
Technology commentator Om Malik has published an analysis challenging Anthropic's approach to naming its most powerful AI model "Mythos," arguing the choice reflects a deliberate shift toward mythological storytelling rather than rational transparency [1]. The name itself, derived from ancient Greek, denotes "a story received rather than verified," according to Malik's interpretation [1].
Malik contends that Anthropic has employed carefully orchestrated public relations efforts—including visits to the White House and Vatican—to establish moral legitimacy for its AI technology while avoiding external scrutiny [1]. He characterizes this strategy as "moral performance" coupled with "controlled revelation," designed to accumulate credibility without subjecting the technology to genuine independent verification [1]. The critic draws historical parallels to Augustus's consolidation of power, medieval monarchy's doctrine of sacred kingship, and the doctrine of Manifest Destiny to illustrate the pattern he identifies [1].
Anthropic has justified restricting public access to Mythos by citing cybersecurity concerns, instead offering limited access through "Project Glasswing" to a small number of trusted institutions [1]. Anthropic CEO Dario Amodei stated in a 60 Minutes interview: "I'm very uncomfortable with these decisions being made by a small number of companies, by a small number of people. Nobody elected me" [1]. Malik's critique suggests this stated discomfort about centralized AI decision-making stands in tension with Anthropic's own operational practices regarding model access and information disclosure [1].