美国安全机构正在使用Anthropic的Mythos Preview模型,该模型具备异常强大的网络攻防能力,目前仅向约40个组织开放访问。然而,Pentagon部分部门却将Anthropic视为"供应链风险"。这种矛盾揭示了美国政府在AI供应商信任问题上的内部分裂:一方面急需Mythos的安全能力,另一方面又对其可靠性存疑。
The U.S. security state is using Anthropic's Mythos Preview even while parts of the Pentagon are arguing that Anthropic itself is a "supply chain risk."
Mythos, per Anthropic, has unusually strong offensive and defensive cyber ability, with access reportedly held to about 40 organizations rather than a public launch.
That creates a blunt contradiction: one arm of government appears to want the model badly enough for security work, while another is still fighting over whether Anthropic can be trusted as a vendor at all.
---
axios. com/2026/04/19/nsa-anthropic-mythos-pentagon