Language AI can democratise access to accountability institutions. Examine with reference to recent initiatives such as BHASHINI's integration with the CAG.

Accountability institutions mostly work in English, but most citizens do not. On 30 September 2026, the Digital India BHASHINI Division (under MeitY) and the CAG launched CAG-BHASHINI Anuvaad, an AI tool that translates official audit documents [1]. It can widen who reads audit findings. But being able to read a finding does not by itself make the government answer for it.

How Language AI widens access

• Reaching citizens: CAG K. Sanjay Murthy said audit findings must reach citizens in languages they understand [1]. BHASHINI covers 36 Indian text languages [1].
• Beyond legislators: Under Art. 151, audit reports go to the President or Governor, who has them laid before the legislature [2]. Translation lets taxpayers, local media and civil society read them too.
• Built for audit, and secure: The tool is tuned for audit terminology and designed for security, controlled access and data sovereignty. Sensitive audit data does not pass through foreign translation services [1].
• Can be copied by other institutions: Sansad Bhashini, set up under a Lok Sabha Secretariat–MeitY MoU, translates parliamentary debates and committee papers [5].

Limits to this promise

• Reading is not action: After tabling, the Public Accounts Committee (PAC) examines audit reports. In Kerala, 157 audit paragraphs were still awaiting PAC examination on 31 December 2024 [3].
• Accuracy risk: Audit findings depend on exact words, such as "irregular" versus "wasteful". Human oversight is kept [1], but it is not clear who checks each language or which version counts if they differ.
• Uneven coverage: BHASHINI supports only 23 languages in voice [1]. Models for low-resource languages (those with little digital text) may carry bias [4]. These are the languages that reports on tribal and remote areas need most.
• Two goals in one tool: The tool was launched at the close of Hindi Pakhwada and sits inside Rajbhasha (official language) workflows [1]. Success could end up measured as Hindi compliance inside offices rather than citizen access.

Way forward

• Accountant General offices should publish human-checked summaries of State audit reports in the State's own language.
• Sansad Bhashini should be extended to PAC reports and Action Taken Notes (the government's replies), so citizens can follow a finding from audit to response [5].
• Each translation should be labelled with how it was checked, and one version should be named as authoritative.

In short, Language AI removes the first barrier to accountability, which is understanding what the auditor found. Combined with faster PAC follow-up and quality checks on translations, it can make the CAG's constitutional role under Arts. 148–151 matter to ordinary citizens. It also supports SDG 16.10 (public access to information).