White House to give US agencies Anthropic Mythos access, Bloomberg News reports
Summary (EN)
Reuters reported on April 16 that the U.S. government is planning to make a version of Anthropic’s frontier model Mythos available to major federal agencies, citing Bloomberg. The move would come through a controlled process led by the White House Office of Management and Budget, which is reportedly working with model providers, industry partners, and the intelligence community to establish safeguards before any release. Reuters said the development follows Anthropic’s April 7 introduction of Mythos under Project Glasswing, a restricted initiative allowing select organizations to use the unreleased Claude Mythos Preview model for defensive cybersecurity purposes. The model has drawn unusually intense attention because experts say its high-level coding and vulnerability-finding capabilities could materially change cyber offence and defence. According to Reuters, Mythos has already been credited with identifying thousands of serious vulnerabilities across software categories such as operating systems and browsers. Gregory Barbaccia, the White House federal CIO, reportedly told cabinet officials that protections were being set up to allow agencies to begin using a modified version of the system, though the email did not specify a final timeline or concrete deployment plan. The event is important because it signals a possible shift from frontier-model containment toward selective government operationalization, while also underscoring the degree to which cyber-capable AI has become a national-security and public-sector infrastructure issue rather than solely a commercial product matter.
Summary (ZH)
路透社于 4 月 16 日援引彭博报道，美国政府正计划向主要联邦机构提供 Anthropic 前沿模型 Mythos 的一个版本。该计划据称将由白宫管理和预算办公室牵头推进，并在正式开放前与模型提供方、产业合作伙伴以及情报部门共同建立相应的保护措施。路透社指出，这一进展发生在 Anthropic 于 4 月 7 日通过 Project Glasswing 受限推出 Mythos 之后。按照该项目安排，只有少数获准机构可以出于防御性网络安全目的使用尚未公开发布的 Claude Mythos Preview。之所以引发高度关注，是因为多方专家认为，该模型强大的编码与漏洞发现能力，可能显著改变网络攻防的能力边界。报道提到，Mythos 已被认为在操作系统、浏览器等软件类别中发现了数千个严重漏洞。白宫联邦 CIO Gregory Barbaccia 据称已向内阁部门官员发信，表示相关防护体系正在建立，以便未来允许机构使用一个经过修改的版本，但邮件并未明确最终时间表或具体部署方式。这一事件的重要性在于，它显示美国政府正考虑把高度敏感的网络安全型前沿模型，从受控实验逐步引入政府运营体系，也表明具备强网络能力的 AI 已不再只是商业产品议题，而正在上升为国家安全和公共部门技术治理问题。