\n\n
Key developments: Recent policy discussions and guidance stress that AI procurement must go beyond traditional software safeguards to address the risk of lock‑in specific to AI models and their harnesses. Indian guidelines for cloud services are being extended to cover AI, mandating data ownership and migration standards. Experts argue that procurement rules should require the ability to swap both model providers and the surrounding infrastructure, favouring open‑weight models while still ensuring workflow and integration portability. A proposed shared inference platform would let agencies source compute from multiple approved vendors, boosting competition. New contract language is being recommended for AI purchases, adding clauses that prohibit unwanted training on agency data, require audit rights, enforce transparency on model changes, and set rapid incident‑notification windows.
Why it matters: Without these safeguards, governments risk becoming dependent on single AI vendors, limiting flexibility, increasing costs, and exposing sensitive data to uncontrolled model updates. Enforcing portability and audit rights ensures agencies retain control over their data and can adapt to evolving technology or policy shifts without costly re‑engineering.
What to watch next: • Implementation of shared inference marketplaces in pilot agencies; • Adoption of the expanded Indian procurement guidelines and similar frameworks in other jurisdictions; • Monitoring of contract clauses in upcoming AI vendor agreements to see if the recommended safeguards become industry standard; • Possible regulatory actions that formalise incident‑reporting timelines for AI systems used in the public sector. [[1]]