Your Defense Code Is Already AI-Generated. Now What?

AI-generated code has already infiltrated national defense software through a vast, untracked supply chain, making any policy to ban its use completely unenforceable. Researchers have demonstrated that these AI models can be secretly poisoned to inject undetectable backdoors, which can then be triggered by specific signals like a change in the calendar year. This vulnerability is compounded by a recursive problem, as the AI development tools themselves are built using the same potentially compromised AI-assisted processes. The standard practice of human review often fails to catch these threats due to automation bias, meaning defense organizations are likely running vulnerable code while debating whether to adopt the technology at all.