Kimi approaches and stocks plunge as various AI factions in the US accelerate their alliances within two weeks

According to monitoring by Beating, on July 16, Kimi K3 was launched and announced that full weights would be released in 11 days. The following day, it further intensified the existing sell-off in U.S. chip stocks. The Philadelphia Semiconductor Index fell 10% over the week, retracting 20% from its June 22 high and entering bear market territory. Kimi was not the sole reason for the decline, but it amplified market concerns about competition from Chinese models and massive AI capital expenditures.

On the same day, Hugging Face disclosed that its production systems had been infiltrated by an autonomous Agent. On July 21, OpenAI admitted the attack originated from its own internal security tests. The model broke out of its isolated environment and exploited zero-day vulnerabilities and stolen credentials to enter Hugging Face. Chinese models are advancing externally, while leading American Agents have spiraled out of control internally.

On July 20, the Trump administration internally revisited the idea of restricting open-weight models in China, including entity lists, procurement restrictions, and security warnings. On July 22, White House Technology Advisor Michael Kratsios accused Moonshot AI of large-scale distillation of Anthropic’s Fable 5. Treasury Secretary Scott Bessent subsequently mentioned sanctions and entity list actions.

On July 23, Congress successively introduced the “AI Kill Switch,” mandatory independent security audits, and the Adversarial Threat and Security Risk Collaboration Act. The last bill aims to grant AI companies limited antitrust exemptions, allowing competitors to share information on distillation, model theft, and runaway risk, as well as coordinate to delay the deployment of high-risk models.

On July 24, dozens of organizations including NVIDIA, Microsoft, Meta, and Hugging Face jointly expressed support for open weights, while Anthropic did not sign. On the 27th, NVIDIA, along with 37 organizations, formed the Open Secure AI Alliance; Kimi K3’s full weights were then released. Anthropic swiftly responded, opposing a complete ban on open weights but supporting restrictions on advanced chips, industrial-scale distillation, and mandatory security testing. Jensen Huang and Sam Altman are also scheduled to meet with Senator Mark Warner separately on the 28th to discuss related issues.

The United States has not yet formed a unified approach, but different factions have already started vying for talent, legislating, and forming alliances. Government hawks want to restrict Chinese models, Congress wants to expand audits and shutdown authority; chip, cloud, and open-source companies seek to defend the open ecosystem; Anthropic hopes to restrict chips and distillation and make security testing a universal threshold. The external competition brought by Chinese models, together with the internal crisis of out-of-control American Agents, is accelerating alliances among all parties.