Summary

  • GAO found that none of 24 major federal agencies had tested post-quantum cryptography in their own environments during its February 2024–September 2025 audit. One had researched vendor candidates; the other 23 had not prepared to test.
  • OMB's 2022 guidance called for inventories, funding assessments and testing. NIST finalised core standards in August 2024. The operational gap lies between a national migration instruction and proof that new cryptography works in specific legacy systems.

A finished standard meets an unfinished system

The GAO's October 2026 public report describes a peculiar asymmetry. The federal government can specify post-quantum algorithms, ask agencies to count vulnerable systems and request spending estimates. It cannot point to a test of those algorithms in any of the 24 agency environments GAO reviewed. The audit ran from February 2024 through September 2025; its findings describe that period. The public version was released on 6 October 2026 and clarified on 14 October. It offers no fresh November 2026 census of agency tests.

NIST finalised its first core post-quantum standard in August 2024, alongside two digital-signature standards. The mathematics is a necessary start. The difficult operational questions live elsewhere: which application calls which cryptographic library, whether a larger key or signature fits a protocol field, what a supplier has shipped, how an older device receives an update and who owns the rollback if authentication fails. Those questions cannot be answered by admiring a specification.

OMB's November 2022 direction asked agencies to prioritise inventories of vulnerable cryptography, assess funding needs and test post-quantum cryptography in agency environments. On testing, one agency had done market research into possible vendor implementations, but had not performed the test. The other 23 had not addressed either test-preparation activity GAO assessed. The result was zero tested environments among 24 agencies during the audit.

The word “voluntary” did real work

Sixteen agencies told GAO it was too early to identify candidates or test because vendors were still integrating the standards into products. Officials at the Office of the National Cyber Director called agency testing voluntary and cited limited resources. GAO pushed back on both points. OMB's guidance did not describe testing as voluntary, and an agency could prototype with open-source tools before a complete commercial catalogue appeared. The disagreement is an institutional one: does a central plan require a local demonstration, or can everyone wait for the market to provide one?

A bounded prototype would not solve a government-wide migration. It would expose real compatibility limits before they are hidden inside a budget estimate. A test in one agency system can show which component must change, who controls it and whether a vendor roadmap is a dependency or an excuse. The absence of all 24 tests is striking because even a small experiment would have produced more information than a conference call about readiness.

The upstream paperwork was incomplete as well. Only one agency fully addressed the applicable inventory activities. One had no priority-system inventory and 22 had partial inventories. GAO checked vulnerable-algorithm accuracy in a selected six-agency sample; none fully met that specific accuracy test. It would overstate the evidence to apply that six-agency result to all 24. The broader count already tells the central story: most agencies had no complete map from priority system to cryptography needing replacement.

A rough price on an unknown job

None of the 24 agencies fully addressed the funding-assessment activity. Twenty-one gave partial assessments with acknowledged inaccurate data; three gave none. Just one funding assessment rested on a complete priority-system inventory. OMB and the National Cyber Director's office described the estimates as rough orders of magnitude for policy decisions, rather than appropriation requests. That description is fair. It also makes the sequence plain: without a reliable inventory and a test of a real system, a migration price is largely a placeholder for work nobody has yet measured.

The complete GAO report names the consequence without claiming a present catastrophe. Today's quantum machines cannot break the relevant cryptography. A future sufficiently capable machine could, and encrypted material captured now may still matter when that capability arrives. The credible present-day risk is a long delay in finding and replacing vulnerable dependencies, especially for data that must remain confidential for years.

GAO made 89 recommendations in its sensitive September 2025 report; the October 2026 public version added none. Counting recommendations is another easy substitute for counting tested systems. The stronger public evidence would be dated, agency-specific inventories; bounded prototypes in actual environments; compatibility results; named owners for legacy blockers; and budgets revised from those results. That evidence would change this audit's conclusion. It was absent from the period GAO examined.

Heng Lu's account of local implementation after a shared initial specification is useful as an analytical lens, not a source of federal facts. The federal standard sets a common starting point. Every agency still has to make a local, testable decision about its own systems. Washington has completed the part that can be printed. The test bench is where the migration begins.

Sources