www.lesswrong.com/posts/PBrggrw4mhgbksoYY/a-tale-of-three-contracts

Multiple credible sources covering the same official spell the name Pete Hegseth (not “Hegsted”). For example, AP’s coverage of his confirmation uses “Pete Hegseth,” and TIME’s coverage of the DOD rebrand to “Department of War” also uses “Pete Hegseth.” This makes the post’s spelling “Hegsted” a factual error.

Note: whether the title is “Secretary of War” vs “Secretary of Defense” appears to depend on the reported rebranding context; the clear error I can verify is the misspelling of the surname.

The post claims Claude Gov is the only LLM deployed on classified networks. But Microsoft’s Azure Government communications indicate their GPT models are available in U.S. Government Secret and Top Secret cloud environments (classified, air-gapped environments). That directly contradicts “only.”

Additionally, reporting from Defense One/Nextgov describes Microsoft making GPT-4 available in the Azure Government Top Secret cloud for classified U.S. government workloads and mentions use via DoD’s classified network (even if operational go-live may require additional approvals). Together, these sources provide concrete evidence that other LLMs besides Claude Gov are deployed/available in classified environments.

This is a broader version of the earlier “only” claim (“only model” rather than “only LLM”). Microsoft’s Azure Government announcement says GPT-5.2 is available in Secret and Top Secret cloud environments (classified, air-gapped clouds). That alone is enough to contradict the statement that Claude Gov is the only model deployed on classified networks.

Defense One’s reporting further supports that Microsoft has also brought GPT capabilities into Top Secret classified cloud offerings for U.S. government workloads.