Pentagon blacklists Anthropic, but the NSA keeps using Claude

The US government has landed in a paradoxical position. The Pentagon has formally placed Anthropic on a national security blacklist as a supply chain threat, yet the intelligence community and the NSA are authorized to keep using the company's advanced models. The reason is simple: the government has no workable alternatives.

How the problem arose

Anthropic came under fire in 2025 for refusing to work with defense contractors. That was defensible from an AI ethics standpoint, but it created a serious dilemma for the US government. After the ban on chip exports to China and restrictions on access to computing capacity, the US faced an acute shortage of reliable software for deep data analysis. Anthropic controls a significant share of this market through its flagship model, Claude.

The Pentagon formally put the company on a sanctions list, declaring it a threat to the national security supply chain. On paper this looked like an attempt at applying pressure, but in practice it created an unforeseen mess.

An exception born of necessity

The intelligence community could not simply give up the best tool on the market. The NSA and affiliated agencies kept using Claude for critical data work, and no blacklist changed that reality. As a result, the White House had to authorize a special exception.

Chief of Staff Susie Wiles approved the NSA's continued use of Anthropic's advanced model despite the company's official status. This rare and plainly awkward decision reveals the real priority: when national security is at stake, pragmatism beats politics.

Government sources unofficially acknowledged the core of the problem:

• There is no real alternative to Claude among frontier-level models on the market
• The ban hurts intelligence work more than the hypothetical threat from Anthropic
• Geopolitical pressure works only if a workable replacement exists
• Claude is embedded in critical analytics systems
• Switching away would require years of retraining and rewriting of tools

A lesson for the AI industry

This paradox illustrates a basic rule of technology geopolitics. Ideological bans on companies work only under competition and when alternatives exist. If one vendor monopolizes access to the best tool, sanctions become a nice gesture at the diplomatic level.

Paradoxically, this works in Anthropic's favor. Its refusal of defense contracts was seen as an ethical stance, but in practice it created a monopoly. The government is officially against the company yet unofficially depends on it, which is more powerful than any contract.

What this means

In the long term, this undermines stability. The US is investing in developing its own models, but as of 2026 there are no results yet. As long as Claude remains the standard in intelligence, any complications with Anthropic will create a threat to national security that the US will have brought on itself.