Washington has a tangled AI strategy. If you can’t control the technology, ban the vendor. If employees keep using it, call another meeting.
Anthropic’s showdown with the Pentagon over autonomous weapons and mass domestic surveillance exposed a deeper problem. Washington can fight over which AI companies get government contracts. It can’t stop federal employees from wanting AI that works.
Enter Shadow AI. Rep. James Walkinshaw (D-VA) flagged the emperor’s naked issue.
Federal employees want to analyze documents, write code, summarize intelligence, and escape the bureaucratic obstacle course. If approved tools fall short, people will find alternatives. Personal accounts. Unapproved applications. Workarounds that make the job easier and the security officer’s blood pressure spike.
That’s not an argument for ignoring security. It’s an argument for getting serious about it.
Banning a vendor doesn’t eliminate demand. It can push AI use underground, beyond the visibility of agency security teams. Congratulations, Washington. You’ve won the procurement battle and potentially lost the data.
The government needs clear rules, secure alternatives and visibility into how employees actually use AI. It also needs to distinguish legitimate security concerns from the illusion that prohibition equals control.
Here’s the punchline: Shadow AI thrives wherever official AI policy meets unofficial workplace reality.
Washington can ban a product. It can’t executive-order away demand.
Until agencies deliver secure AI tools that people actually want to use, Shadow AI will keep working overtime. And nobody will have the contract to manage it.