DeepSeek, a Chinese-owned company, recently released an AI chatbot that has caught the attention of legal tech vendors. They see potential in the model due to its open-source nature, reduced cost and customization possibilities.
Launched in January, DeepSeek's AI chatbot gained traction due to its efficient performance that is on par with other models, such as OpenAI's ChatGPT. It also uses a different approach to reasoning, which is how an AI chatbot uses logic to analyze information.
DeepSeek's appeal is that it is a lower-cost model that keeps up with the major generative AI providers, and it is an open-source option for organizations that want to deploy it, according to Joseph Regalia, an associate professor of law at the University of Nevada, Las Vegas, William S. Boyd School of Law.
"That is what everyone is after in the legal world — a GenAI that is powerful enough to make a big difference but controllable," Regalia told Law360 Pulse. "In many ways, that is the GenAI holy grail for law firms."
Weighing DeepSeek's Cost Savings
Cost is a big reason why some are bullish on DeepSeek.
Most generative AI tools, including ChatGPT, can be costly because of the development of complex models and the hardware to process the data. AI models also run on units of data known as tokens, which are also expensive for vendors and law firms.
Some law firms argued that generative AI tools can be too expensive, as the savings from reducing lawyer workloads might be offset by AI fees.
DeepSeek is considered more cost-effective than other AI models because the development expenses are lower, and it is more efficient with graphics processing units. DeepSeek now offers a leaner option, according to Praful Saklani, CEO of the AI contract management system Pramata.
"I fully expect that price per token and the performance availability [will] drop by a couple of orders of magnitude over the next 12 months," Saklani said. "If we're paying the same amount that we are today, I'll be shocked a year from now."
Saklani added that AI-based unit costs should be more manageable going forward, resulting in more affordable AI tools created by law firms and vendors.
At least one legal tech vendor has already fully adopted DeepSeek.
In February, the contract intelligence provider Icertis integrated its platform with DeepSeek to dive into contract data and enable enterprises to achieve greater cost control over AI.
"Our integration with DeepSeek exemplifies the Icertis OmniModel commitment to help customers capitalize on the latest advancements as the AI landscape evolves, thereby accelerating their journey toward contract intelligence and maximizing the value of every customer and supplier agreement," said Sudarshan Chitre, senior vice president of artificial intelligence at Icertis, in a statement.
Pramata is also experimenting with DeepSeek's open-source model to see how it might fit in with its platform, according to Saklani.
Overcoming DeepSeek's Risks
John Tredennick, CEO and founder of Merlin Search Technologies, wrote in a January LinkedIn post that DeepSeek "signals a fundamental shift in how AI systems can be developed, deployed, and accessed" in the legal profession. He said DeepSeek can lower the operating costs of AI and can reduce latency to improve real-time applications such as document review.
However, Tredennick warns that Deepseek carries security and compliance risks.
Using DeepSeek's chatbot poses several significant risks, as the company stores data on servers in China, which means potentially exposing confidential information to the Chinese government. There are also concerns about weak data security protections on the platform.
Some law firms have called the platform a "privacy nightmare" for safeguarding client data. Saklani said he wouldn't recommend any organization use DeepSeek's AI chatbot because of these risks. The solution to mitigate such risks is for legal teams to use an open-source version of the model through provisioning on a trusted cloud platform with strong security and privacy standards.
"You can provision a model of DeekSeek that is fully locked out to the rest of the world," Saklani said.
UNLV's Regalia said that while he doesn't expect law firms to use DeepSeek directly because of the data privacy risks, he anticipates organizations will deploy DeepSeek's open-source AI models internally so that firms can maintain control over their data and customize the models to fit specific legal domains.
Merlin Search Technologies, which produces software for investigations and discovery, confirmed to Law360 Pulse this month that, if it added DeepSeek's model, it would be through Amazon Web Services, or AWS, to harness it safely and protect data.
AWS announced in January that DeepSeek's model is available through the Amazon Bedrock Marketplace. Microsoft also announced that DeepSeek's model is available on Azure.
For law firms still uneasy about DeepSeek's open-source model, Saklani expects "copycats" from places with stronger legal protections, such as North America and Europe, to release similar low-cost AI models.
Saklani said that DeepSeek is a "turning point" in the legal industry, as vendors begin to test this new AI model. It is also a signal to the broader industry that there will be open-source options for legal teams that can be more affordable and customizable.
"Plan for a world where the GenAI capability will be ubiquitous and will be very cost-effective," Saklani said. "The journey is clearly pointing in that direction."
--Additional reporting by Sarah Martinson. Editing by Nicole Bleier and Dave Trumbore.
For a reprint of this article, please contact reprints@law360.com.
