Anthropic can serve its models within the security standards required to handle classified data. The other labs do not yet claim to have this capability.

Even if they do, I assume the other labs would prefer to avoid drawing the ire of the administration, the public, or their employees by choosing a side publicly.