With legal AI firm Harvey setting up shop in Bengaluru, we speak with CEO and co-founder Winston Weinberg and CTO Siva Gurumurthy about India’s place in the company’s global growth plans. By Chandu Gopalakrishnan
when Harvey CEO and co-founder Winston Weinberg left a promising career as a litigator to start the legal AI company, Harvey, he was not just another lawyer dabbling in tech. He and co-founder, Gabe Pereyra, an AI researcher who previously worked on large language models (LLMs) for Google Brain and Meta, were convinced that generative AI could transform the entire profession.
Three years later, Harvey has grown from a scrappy idea into one of the fastest-rising companies in legaltech, with more than 500 enterprise customers, unicorn status at a USD5 billion valuation, and a client list including some of the world’s most powerful law firms and financial institutions.
Weinberg’s big break came in 2023, when Allen & Overy rolled Harvey AI out to thousands of lawyers across 43 offices. Since then, he has doubled down, teaming up with Microsoft and LexisNexis to create tools such as ContractMatrix, and pushing into what he calls “agentic AI”: systems designed to automate complex, multi-jurisdictional tasks that normally consume senior associate hours.
Now Harvey is setting its sights on India. The legal services market in India is projected to reach USD45.2 billion in revenue in 2024, and USD67.4 billion by 2030.
The company is opening a Bengaluru office just as the country’s legal services sector reaches a crossroads.
“Our in-house Privacy and AI Legal team closely monitors new developments and enforcement timelines …to ensure we are ready to implement any additional requirements the government may introduce, including data transfer mechanisms.”
The changing outlook
Harvey AI already counts marquee names including AZB & Partners, Shardul Amarchand Mangaldas & Co, and S&A Law Offices on its Indian client roster. Announcing its global rollout of Harvey AI in September, AZB & Partners said it would help them “near-automate routine tasks” such as document review and translation.
“We believe that generative Al is not just a tool, but a transformative force that will empower our lawyers to deliver deeper insights, faster results, and even greater value to our clients,” says Zia Mody, co-founder and managing partner of AZB.
Harvey is also working with enterprise companies in India. “In India, the possibility for both in-house work and firm transformation is notably high,” Weinberg tells India Business Law Journal.
A recent IBLJ report showed paralegal roles under pressure from automation, but Weinberg frames it differently: “We generally believe that AI won’t replace lawyers; it will replace lawyers who don’t adopt technology as part of the broader transformation of law.”
He says that Harvey AI is supporting reskilling through its law school alliance programme in the US, partnering with prestigious institutions such as Stanford Law School, NYU School of Law, the University of Michigan Law School, UCLA School of Law, the University of Texas School of Law and Notre Dame Law School as collaborators, after beginning “a rigorous pilot of Harvey’s platform a year ago”.
Lawyers in the US, Canada and some Caribbean courts are now required to disclose the use of generative AI tools in preparing court documents. US federal courts, such as those in Texas, Pennsylvania, Illinois, and the Court of International Trade, mandate disclosure and certification that AI-generated content has been verified for accuracy.
Canadian courts, including Manitoba’s Court of King’s Bench and the Supreme Court of Yukon, have similar disclosure requirements. Courts emphasise that AI cannot be used to generate affidavits, witness statements or evidence, and lawyers remain fully responsible for their submissions. Failure to disclose or verify AI-assisted documents may lead to penalties, rejection of filings, or disciplinary action.
Deloitte recently had to partially refund the Australian government after delivering a AUD440,000 (USD289,000) report that contained multiple errors attributed to the use of generative AI. The report, commissioned by the Department of Employment and Workplace Relations to review the targeted compliance framework and related IT systems in the welfare programme, included fake academic citations and a fabricated quote from a federal court judgment. After academics flagged these errors, Deloitte acknowledged using a generative AI model in early drafts.
In India, while mandatory disclosures are not yet widespread, a growing trend influenced by US court practices is evident, and the Supreme Court of India has been cautiously exploring AI applications in legal processes.
Weinberg remains optimistic that more players in India’s legal sector will see AI-assisted operations as an opportunity. He wants law firms and general counsel to move from being customers to building custom workflows with Harvey, “to integrations and partnerships that help drive transformation” in the Indian legal services market. “I hope what GCs and managing partners read into our expansion plan is that Harvey plans to build a significant engineering team in India, and that as a result there are countless opportunities for how they can engage with our team,” he says.
Bengaluru also offers the chance to establish a strong tech back office. “We are starting [in Bengaluru] with a focus on engineering, product and design talent led by Siva Gurumurthy, our CTO,” says Weinberg. “Siva has built teams in India and was a key driver in our decision to build a team there now.
“We’re hiring the best possible engineering talent to focus on the most important AI product challenges that will help our customers, so I anticipate their work will help us accelerate growth in India, but also well beyond India, which I hope is one of the many reasons people choose to work for us.”
Winston Weinberg, left, with co-founder Gabe Pereyra
Changing compliance norms
Harvey’s entry into Indian legaltech comes at a time of great transition. The Digital Personal Data Protection Act, 2023 (DPDP Act), has been passed but is not yet fully in force, and norms for how foreign firms must manage India-resident data processing remain unclear. Weinberg is well aware of the implications.
“Our in-house Privacy and AI Legal team closely monitors new developments and enforcement timelines, while working with leading local counsel to ensure we are ready to implement any additional requirements the government may introduce, including data transfer mechanisms,” he says.
For data localisation, Harvey already offers customers to store their data exclusively on Microsoft Azure servers in the central India region. According to Weinberg, customers act as the data fiduciaries and can specify storage preferences in their order form, ensuring data remains within India.
Under the current interpretations of the DPDP Act, a data fiduciary is an entity or person who decides the purpose and means of processing personal data. Harvey AI functions as the data processor, handling personal data on behalf of the fiduciary.
In the Indian context, we are making commercially reasonable efforts this year to incorporate local primary-law sources such as SCC Online, Manupatra, etc., into the platform. These integrations will ensure that users have access to authoritative material directly within Harvey
“To safeguard personal data, Harvey maintains a layered security program aligned with international standards,” says Weinberg.
“Our Security Addendum details these controls, which include encryption in transit and at rest, logically isolated customer environments, strict identity and access management, continuous vulnerability monitoring, and independent third-party penetration testing.”
AI, research and quality
Along with data processing comes the crucial issue of data input. Generative AI products rely heavily on broad contextualisation of information, whether provided directly by users or sourced freely online.
According to a Thomson Reuters Westlaw study on cases between 30 June and 1 August 2025, hallucinations and citations of non-existent legal cases remain pervasive across courts.
“This search found 22 different cases in which courts or opposing parties found non-existent cases within filings, leading to discipline motions or sanctions in many instances,” writes Zach Warren, a manager for enterprise content for technology and innovation at the Thomson Reuters Institute.
Whether integrating with local primary-law sources or drawing on customer-provided corpora alongside their LexisNexis alliance, the responsibility for demonstrating citation reliability in Indian courts and tribunals lies with Harvey AI.
“In the Indian context, we are making commercially reasonable efforts this year to incorporate local primary-law sources such as SCC Online, Manupatra, etc., into the platform,” says Weinberg. “These integrations will ensure that users have access to authoritative material directly within Harvey.”
Unique challenges
The pace and character of legaltech and AI adoption in India are distinct from the West, and in many ways “uniquely promising”, says Weinberg.
“In Western markets, adoption is shaped by mature infrastructures, structured procurement processes, and incremental efficiency gains,” he says. “In India, adoption follows a more dynamic and leap-frogging trajectory.”
He sees three factors that are driving India’s outlook:
-
- Client demand is shifting. Customers now expect firms to use AI for efficiency, reversing earlier restrictions;
- Customisation matters. Beyond generic tools, firms want tailored workflows built for their needs; and
- Partner-level adoption is rising. AI use is no longer left to junior associates; senior partners are driving adoption.
In India, both firms and in-house teams are open to experimenting with AI not only to refine workflows, but to rethink them entirely.
Lawyers look for practical, immediate value, favouring tools that improve turnaround time, research quality and client service. Local expertise is critical; solutions must reflect Indian courts, drafting styles and regulatory frameworks, notes Weinberg.
“What excites me most about India is the openness to new ways of working; young lawyers and forward-looking firms are eager to experiment with AI to differentiate themselves in a competitive market,” he says. “This means innovation in India isn’t just about keeping pace – it’s about reimagining the practice of law with AI at its core.”
Putting Harvey to work
As Harvey sets its sights on India, CTO Siva Gurumurthy shares his perspective on the company’s engineering strategy, data safeguards, lawyer training, and the wider impact of AI on legal practice
Q: What challenges come with localising Harvey for Indian legal workflows? And will the Bengaluru team handle language quality assurance, document management system patterns, or India-specific features?
Siva Gurumurthy
Siva Gurumurthy: The Bengaluru team will certainly work on projects that help Harvey grow in India, but I think it’s important to note that they will work generally on critical projects for Harvey that help our customers globally.
So, I think for really strong engineering talent, working at Harvey gives you the opportunity to solve hard and interesting technical problems for customers at a global scale from India, which is part of the reason we are really looking forward to hiring there.
Q: How do you balance client corpora with proprietary datasets and alliances like LexisNexis? What safeguards ensure accuracy and bias mitigation for India?
Gurumurthy: A key feature of Harvey is its ability to leverage a wide range of public and proprietary legal data sources. This data is used as context for model responses, but importantly is never given access to client-provided data that is stored on the platform.
With Lexis specifically, Harvey is able to search Lexis case law (and other data), and use it to generate a response to a user’s query, but this interaction is one-directional (Harvey reading data from Lexis); at no point does Lexis gain visibility or access to client data stored in Harvey.
As we expand our data sources (which happens weekly), we will follow a similar approach with other data partners. With regard to model accuracy, we share regular updates on model accuracy specific to the legal field. Over time, it’s possible we consider more regional analysis globally, so that’s a good item for us to get feedback from customers in India on.
Q: What skills should Indian lawyers build to stay relevant in an AI-augmented environment? How can firms and universities help?
Gurumurthy: I think it’s less technical skills per se and more developing AI fluency. For example, we see major changes in results when lawyers get good at prompting, so even just getting comfortable with prompts is a great way for lawyers to increase their comfort level with AI products.
At universities, we have just announced a programme in partnership with law schools that will help address this specific issue over time. We announced the first batch of schools in the US, but over time, the plan would be to bring this type of content to any institutions teaching law, so that AI becomes part of what people learn alongside traditional elements of law.
Q: How will AI affect junior associates and paralegals in India? Will Harvey substitute or augment their work?
Gurumurthy: We have seen countless examples of customers whose use of Harvey meaningfully augments the work of lawyers and legal teams more broadly, saving them hours of time every week. We think of AI as improving the daily work of lawyers by making their jobs more efficient and effective.
Q: What cybersecurity protocols protect sensitive legal data in Harvey? How do you address confidentiality and regulatory concerns?
Gurumurthy: As you might imagine, for law firms, these concerns are paramount, so Harvey spends a significant amount of time and energy thinking through these challenges and providing context on them to our customers. To briefly summarise, as best I can, I think it comes down to three things:
(1) Model usage. It adheres to a strict policy of no training, no retention, and no human review;
(2) Complete segregation. All customer data is stored in a dedicated, logically separated storage account with its own unique encryption key. Data is encrypted at rest using AES-256 and in transit using TLS 1.3 (cryptographic methods to protect data); and
(3) Full customer control. Customers control where their data is stored and processed and can set their own data retention policies, including a zero-day retention option where queries and responses are not stored after processing.
On regulatory evolution, we work closely with our head of privacy to ensure we regularly update our policies in each of the markets we operate in, so compliance is always top of mind for our team.