AI code review missed critical vulnerabilities, leading to COLDCARD being exploited last week

Bitcoin News posted on the X platform stating that Coinkite claims the vulnerability exists at the boundaries of two unrelated firmware submodules, rather than in its Bitcoin or crypto code, allowing it to evade manual and AI-assisted code reviews for years.

Coinkite stated that after the incident, the company tested cutting-edge AI models including Kimi K3, Claude Fable, and Codex 5.6, none of which identified the flaw. Coinkite is currently urging security-critical projects to specifically audit build systems and submodule boundaries, warning that AI-assisted development may leave similar blind spots in the Bitcoin ecosystem.