Independent control infrastructure for AI means the tools for testing, monitoring and, if necessary, stopping AI systems are run by neutral third parties, not by the companies that build them. It would cover independent evaluations, usage and behaviour monitoring, compute oversight, and enforceable intervention such as revoking access or shutting a system down. The rationale is that developers face commercial pressures and their self-assurances are hard to verify, while credible external checks build public trust. The main challenges are gaining access to proprietary systems, coordinating internationally, and keeping pace with the technology.