U.S. banking institutions are being encouraged by Trump administration officials to test Anthropic’s Mythos artificial intelligence model, according to reports, even as the Department of Defense has recently designated the AI safety company as a supply-chain risk. The apparent contradiction underscores mounting tensions within the U.S. government over artificial intelligence development, regulation, and the role of private technology companies in critical infrastructure and national security.
Anthropic, a leading AI research company founded in 2021 by former members of OpenAI, has emerged as a significant player in the competitive generative AI landscape. The company has developed large language models including Claude, which has gained traction among enterprise clients. The Mythos model represents its latest advancement in AI capabilities. However, the Pentagon’s recent classification of Anthropic as a supply-chain risk represents a substantial headwind for the company, signaling concerns within the defense establishment about potential vulnerabilities, data security, or foreign entanglement in the company’s operations or technology stack.
The divergence between Pentagon caution and apparent administration encouragement of bank testing creates an unusual policy environment. Supply-chain risk designations typically reflect concerns about vendor reliability, security posture, or geopolitical vulnerabilities. When a company receives such a designation, federal agencies typically restrict procurement and partnerships, and the designation can influence private-sector decisions through signaling effects. Banks, which operate under extensive federal oversight and stress-test requirements, are particularly attuned to regulatory signals about acceptable technology partnerships. The mixed messaging from different branches of the Trump administration may therefore create uncertainty about the appropriate course of action for financial institutions.
The banking sector’s interest in advanced AI models reflects broader industry trends. Financial institutions increasingly deploy machine learning for fraud detection, risk assessment, trading optimization, and customer service. Access to state-of-the-art large language models could enhance these capabilities. However, banks are also heavily regulated, subject to data protection requirements, and must demonstrate that third-party technology vendors meet security and compliance standards. A Pentagon supply-chain risk designation complicates this calculus, as banks must weigh potential competitive advantages from testing cutting-edge AI against regulatory and reputational risks associated with partnering with flagged vendors.
The episode reflects deeper fractures in U.S. AI policy. Different agencies—Defense, Commerce, Treasury, and others—have varying interests in AI development. The Pentagon focuses on military applications and technological advantage against adversaries. The Commerce Department and banking regulators emphasize innovation and competitiveness. Treasury and intelligence agencies prioritize financial system security and counterintelligence. These institutional interests do not always align, and the Trump administration has signaled both deregulatory intent and nationalist competition with China. This creates environments where different officials may send conflicting signals to the private sector about which companies, models, and partnerships are encouraged or discouraged.
Anthropic’s position in this landscape is particularly fraught. The company has positioned itself as AI-safety focused, emphasizing responsible development and transparency. It has no known foreign government connections and is U.S.-based. Yet it operates in an extraordinarily competitive and strategically sensitive domain. Any Pentagon designation as a supply-chain risk likely reflects classified concerns or information not disclosed publicly. The Pentagon may have identified technical vulnerabilities, questioned the company’s security practices, or assessed that Anthropic’s technology could be compromised or stolen. Conversely, some officials may believe the Pentagon’s assessment is overstated or that testing Mythos in controlled banking environments poses acceptable risks while yielding valuable intelligence about AI capabilities.
The situation creates practical dilemmas for banks. Financial institutions cannot easily ignore either Pentagon warnings or administration encouragement. They must perform independent risk assessments while operating in an environment of conflicting official signals. This uncertainty may slow adoption of Mythos in the banking sector, even if the model’s technical capabilities are competitive. Banks may instead gravitate toward AI models developed by vendors without Pentagon risk designations, or wait for clarification on official government policy. Anthropic faces a narrowing window to demonstrate that the Pentagon’s concerns are misplaced or that the company has addressed underlying vulnerabilities.
Looking ahead, the dynamics surrounding Anthropic’s Mythos and the broader AI policy environment warrant close monitoring. Congress may demand clarification from Pentagon and administration officials about AI vendor designations and their consistency with stated policy objectives. The SEC and banking regulators may issue guidance clarifying expectations for bank AI partnerships in light of supply-chain risk designations. Anthropic itself may pursue government contracts, security certifications, or remediation efforts to address Pentagon concerns. The resolution of this particular case will likely set precedent for how other AI companies and other sectors navigate mixed regulatory and political signals in a period of intense geopolitical competition over artificial intelligence.
