Trump Officials Discuss Anthropic Security Risks
Informed sources told Bloomberg that Commerce Secretary Howard Lutnick and Pentagon Chief Technology Officer Emil Michael held a video conference with Anthropic Chief Computing Officer Tom Brown to discuss artificial intelligence security risks. Brown has become the company's face in Washington. The company declined to comment, and the White House did not respond.
The meeting occurred after CEO Dario Amodei published an approximately 3,800-word article on Saturday, reiterating calls for government regulation and control over the pace of advancements, which has strained recently eased relations. Amodei also stated at the Salesforce conference that the economic value of AI is currently only about 5% to 10% utilized, and discussed the company's expansion of model access.
The conflict dates back to this year. In March, the Pentagon labeled Anthropic as a supply chain risk, with Trump calling it "far-left." In June, Lutnick sent a letter requesting that Fable 5 and Mythos 5 obtain Commerce Department approval for all foreign nationals at global destinations, or face civil and criminal penalties, leading to a temporary shutdown of the model for all users. Brown and policy head Sarah Heck subsequently took over negotiations with the Commerce Department and Cyber Director Sean Cairncross. In August, a federal judge overturned the supply chain blacklist, ruling it constituted retaliation; the military stated that appealing was more important and claimed to have moved about 90% of classified AI loads out of the company.
Earlier this month, Lutnick stated, "They have changed," and "We trust Anthropic," introducing Brown to ministers at the G20 Innovation Ministers' meeting. Michael publicly dismissed the doomsday warnings from departing researchers as fear of data centers and insisted that contracts must cover "all legal uses." Anthropic has drawn two red lines: no mass surveillance of Americans and no autonomous control of lethal weapons. Both sides have different narratives regarding the reasons for the breakdown in negotiations.
The video conference linked export control authority with defense procurement authority. Security risks were both Brown's topic and a ticket to reclaim federal computing power and contracts. No public minutes or new licensing or de-licensing texts were released.
In market mechanisms, buyers are federal agencies seeking controllable models for intelligence and logistics, while sellers are laboratories seeking de-licensing of top models and government orders. This is an event-driven relationship repair: the lengthy article reignited opposition, and the computing officer stepped in to shift the conflict from a value dispute to a technical assurance negotiation. The funding flow still connects to competitor models that can access federal loads; the beneficiaries are vendors willing to sign for "all legal uses," while the pressured party is Anthropic, which has made security red lines a product differentiation, and military procurement officials who view the company as a supply chain risk.
Source: Public Information
ABAB AI Insight
Amodei's lengthy article calls for government speed control, while Brown meets with the two individuals who hold that control. This is a division of roles: the CEO seeks a moral high ground externally, while the co-founder seeks computing power and orders. The June export controls demonstrate that the Commerce Department can use the 2018 export reform law to classify cloud models as military items; the March supply chain label shows that the Defense Department can write contract rejections as national security. The judge overturned the label but did not revoke these two powers. Lutnick's mention of "trust" is political language, while Michael's claim that 90% of loads have been moved out is procurement language.
The capital path determines who can sell the most expensive reasoning through federal contracts and export licenses. Anthropic's integration of Claude into Salesforce's 7,000 employees is a hedge on the commercial side; the Pentagon's removal of classified loads is a bloodletting on the defense side. Brown is responsible for rejoining both sides. The subtext of the security talks is: what monitoring and weapon terms must be accepted to regain licenses and orders? Once red lines are written into contracts, they become enforceable product restrictions and can serve as excuses for competitors to seize orders.
This is not ordinary lobbying but rather Boeing negotiating export licenses with the Defense Department while simultaneously discussing foreign users with the Commerce Department. OpenAI and Google signed the "legal use" framework earlier, while Anthropic exchanged two exceptions for the supply chain war. The industry narrative has shifted from laboratory alignment to national machine procurement narrative. Those who can simultaneously handle the Commerce Department's foreign user ban and the Defense Department's command chain language will secure the next round of federal computing budget.
Structural judgments belong to the concentration of licensing rights amid regulatory changes. The mechanism is: model weights exit through interfaces, and regulation manages people through those interfaces. Security risks are a common term in the conference room but translate into licenses and procurement catalogs. The talks themselves do not reduce risks; they only determine who prices the risks.
ABAB News · Cognitive Law
- Security talks are often another table for licensing negotiations.
- The CEO writes lengthy articles, while the co-founder repairs relationships.
- Red lines can serve as branding or as reasons for being moved out.