India, Sept. 27 -- We are no longer just worried about a classified document being pasted into a prompt; we are witnessing the erosion of the state's cognitive perimeter.

A colleague in the government asked me a very interesting question. They wanted to understand the landscape around the usage of commercial AI chatbots in regular government work. That led me thinking and researching, and into a rabbit hole which spanned across official secrecy, informational integrity, Personal data protection, right to information, hallucinated content, model biases and even the sovereignty of the country. Consider the following scenarios. The official types a draft Cabinet note into a commercial chatbot. The provider suffers a breach-not hypothetical,...