US security agency is using Anthropic’s Mythos despite blacklist, Axios reports

Reuters reports the U.S. National Security Agency is using Anthropic’s Mythos Preview AI tool despite a Pentagon supply-chain risk designation against the company. The article highlights growing concern that Anthropic’s latest model could be used to supercharge cyberattacks, given its advanced coding and agentic capabilities. The news is negative for sentiment around AI security and defense-sector oversight, but the immediate market impact appears limited.

Analysis

This is a signaling event more than a fundamental revenue event: the market should read it as the state normalizing use of frontier models despite supply-chain skepticism. That matters because defense procurement usually lags capability adoption by quarters, not weeks, so any loosening of internal restrictions can pull forward budget share toward model providers, secure inference layers, and cyber tooling that integrates agentic workflows. The second-order winner is likely the ecosystem around model hardening, monitoring, and on-prem deployment rather than the base model vendor alone.

The bigger implication is competitive pressure on incumbent cyber vendors: if frontier models materially compress time-to-vulnerability discovery, red-team productivity, and exploit generation, buyers will respond by spending more on detection, identity, endpoint isolation, and zero-trust controls. That can widen the gap between “AI-enabled offense” and “AI-enabled defense,” which is constructive for vendors that can prove lower false positives and faster response automation, but negative for legacy point solutions with high human-in-the-loop costs. The adoption curve could steepen over 6-18 months if public-sector buyers conclude they cannot afford to wait for perfect governance.

Tail risk is a short-cycle cyber incident: a high-profile attack attributed to AI-assisted tooling would likely trigger a near-term regulatory clampdown and procurement pause, especially in the next 1-3 months. Conversely, if no headline breach occurs and the Pentagon keeps broadening model access, the market may gradually re-rate secure AI infrastructure, air-gapped compute, and defense software names over the next 2-3 quarters. The consensus may be overestimating the model vendor’s direct benefit and underestimating the infrastructure and cyber-defense capex wave that follows from sanctioned adoption.

For trading, the cleanest expression is to own beneficiaries of secure deployment while fading unprofitable pure-play model hype. The asymmetric setup is in companies with government trust, compliance layers, and recurring cyber contracts, where incremental AI adoption can lift seat expansion and module attach without requiring perfect demand visibility. Avoid chasing the model platform headline until there is evidence the procurement decision is durable rather than a one-off exception.

Key Decisions for Investors

• Long PANW / short a basket of unprofitable AI model names on any risk-on spike, 3-6 month horizon: thesis is that enterprise and defense spend accrues to secure deployment and monitoring rather than to frontier-model branding; stop if there is a broad public-sector procurement expansion that directly benefits model providers.
• Add to CRWD on pullbacks over the next 1-2 months: if AI-assisted attack capability rises, detection and response budgets should accelerate; target a 12-18 month hold with upside from higher module attach and lower churn risk.
• Buy MSFT as a diversified way to own sovereign/enterprise AI infrastructure demand, but size modestly: 6-12 month view, with the risk that procurement scrutiny slows adoption; best entered on weakness after any cyber-headline volatility.
• Consider a call spread on PLTR into the next 2 quarters if defense AI adoption broadens: the optionality is in workflow integration and government trust, but cap downside with defined-risk structure because contract timing is lumpy.
• Do not chase long-only exposure to frontier model vendors on this headline alone; if holding, hedge with short-dated puts or reduce into strength until there is evidence the Pentagon relationship is stable for multiple procurement cycles.