How AI is making its case in courts, trimming trials and errors

While the Union Ministry of Law and Justice told Parliament in its last Session that Artificial Intelligence was being “explored” in Indian courts, it has been expanding its footprint – from courtrooms to lawyer chambers.

A transcription system called ‘TERES (or Technology Enabled ReSolution)’ is now used during Constitution Bench hearings to convert oral arguments into text in real time, addressing a longstanding bottleneck in recording fast-paced hearings accurately. Developed in India, TERES is also being used by dispute resolution institutions outside the country now, including courts in Dubai, Abu Dhabi and Singapore.

Another AI-enabled system, Adalat AI, operates in High Courts of nine states, and has pilot projects running in five others. In Kerala, Adalat AI’s use is embedded across courtrooms, rather than select benches.

At law firms, AI has taken over tasks from drafting to memos to document review and research.

Deepika Kinhal, the co-founder of TERES, says the system was onboarded through a competitive bid by the Supreme Court in 2023. Their prime mandate was ensuring accuracy, she says. “A wrong or missed word can have serious repercussions.”

Under the TERES’s two-layer system, AI generates the transcript but there are people reviewing and correcting the text in real time.

Court officials say that TERES is essentially used for live courtroom display and for publishing argument transcripts on the Court’s website, and their help in drafting judgments remains a matter of individual judicial discretion.

The Supreme Court is also testing AI-assisted e-filing systems which automatically flag missing documents or incorrect annexures at the filing stage, and running other limited pilots, such as with Jhana AI whose courtroom module for case management is currently used in the Karnataka High Court.

Utkarsh Saxena, the co-founder and CEO of Adalat AI, says their tool helps smooth over small problems that can add up in the lifecycle of a case. “In many courts, witness depositions, cross-examinations and even final orders are still written by hand. Cross-examinations can take weeks to be fully written out, especially when witnesses speak in regional languages,” Saxena points out.

Adalat AI is also being used by judges to manage their workload. Saxena says the system puts cases listed across different stages in a single window, reducing the time spent locating files and context between matters. Transcription is built into this flow and trained in courtroom language, to take it further than a generic AI tool. “When a judge says, ‘Section 128 sub-section 2 clause c read with Section 45’, the system understands the structure,” he says.

In Kerala, where AI is in use at different court levels, the High Court deploys it principally at the filing stage to scrutinise vakalatnamas, pleadings and annexures. Sources said around 46,000 cases have been screened through this process, cutting time lost in correcting mistakes, without manual intervention. The court has also developed its own voice-to-text tools, which are being extended to district courts, along with systems that help judges navigate large case files and summaries.

Another proposal under development is a WhatsApp-based interface through which case status can be accessed.

Arghya Bhattacharya, Adalat AI’s co-founder and Chief Technology Officer, says given the sensitivity of the information at their disposal, security has been built in through three non-negotiables: all systems are deployed within sovereign borders; all AI models are built and deployed in-house without third-party access; and no data can be accessed without explicit consent.

Bhattacharya says Adalat AI has also factored in infrastructure constraints such as unreliable connectivity and power cuts, and prototypes were shown to judges and their feedback taken before any code was written. At the same time, says Saxena, they are clear that their role is of assisting rather than replacing court staff. “The stenographer’s role doesn’t disappear. It changes. The job becomes verification.”

The Supreme Court also deploys a translation platform called SUVAS (Supreme Court Vidhik Anuvaad Software) now, which converts judgments and orders into regional languages. As of March 28 last year, 83,783 Supreme Court judgments had been translated into 18 vernacular languages and uploaded on its portal. Hindi accounted for over 36,000 of the translations, with significant volumes in Punjabi, Bengali, Gujarati, Malayalam, Marathi, Tamil, Telugu, and Urdu.

Research assistance is the third area where the judiciary is turning to AI. Tools such as SUPACE and LegRAA scan large volumes of judgments to extract facts, identify legal issues, and list relevant precedents.

While courts proceed cautiously, AI adoption among lawyers and law firms has been fast and accelerating.

Most legal work falls into three buckets, says Suhas Baliga, founder of Axara AI: drafting, review and research. Of that, drafting was the most fit candidate for AI, as it requires preparing petitions, applications, and contracts that follow set formats. AI helps in producing an initial draft and checking for routine requirements. Document review took longer to gain trust, says Baliga. Lawyers were sceptical, particularly when documents ran into hundreds of pages, but improvements in how models handle large volumes of text have softened that resistance, he says.

The third frontier, research, long seen as unsuitable for language models, has also shifted, Baliga says. “AI was not built for research. But now it can synthesise precedents and reasoning, if backed by good data, search and precise prompting.”

However, pricing of specialised legal AI tools remains a barrier, which means that many lawyers rely on free, general-purpose platforms, making their use constrained. But, Baliga says, “It’s rare to find someone who says they don’t use AI at all. AI is already part of their workflow.”

Listing the advantages of AI, Varun Hemachandran, a Senior Curator at Agami, a non-profit organisation working on technology and policy in the justice system, says: “Most of what slows courts down isn’t the law, it’s access to files, to information, to systems that actually work.” Agami hence has concentrated on easing how judges, lawyers and litigants move through the system, Hemachandran says, with a special focus on testing tools within real courtroom conditions, where time, infrastructure and training are limited.

A tool Agami has designed is JIVA, to improve judicial access to legal texts during hearings, while it has also developed systems to track e-Gazette publications, monitor amendments, and create authoritative repositories. “If a judge refers to a statutory provision, the system should pull up the authoritative text immediately,” Hemachandran says.

Komal Gupta, Chief Innovation Officer at Cyril Amarchand Mangaldas, says the AI journey at one of India’s largest law firms began as early as 2017, when most tools were based on machine learning rather than today’s generative models. “At that stage, the technology depended heavily on the quality and relevance of the data you fed into it,” she says. Many tools were trained on foreign datasets, which often made them ill-suited to Indian legal work.

Now, Cyril Amarchand Mangaldas uses AI across areas, including mergers and acquisitions, disputes, and regulatory work.

In due diligence exercises that once involved manually reviewing thousands of contracts, AI systems now go through large document sets and flag relevant clauses or missing provisions. Where the tool is uncertain, Gupta says, it explicitly signals that ambiguity.

Firms also use tools such as CaseMine, which allow lawyers to find relevant precedents in seconds as well as identify precedents that have been overruled or should not be cited. Other tools, including Legora, integrate directly into MS Word, allowing lawyers to draft and revise documents alongside AI assistance.