Anthropic, Meta, Microsoft, Genentech (Roche Group) and IBM have created a shared commons for foundational artificial intelligence patents, aiming to reduce legal disputes and accelerate development.
The effort could also shape the global race to build more advanced, general-purpose AI systems.
The five founding companies last week announced the formation of the Shared AI License Foundation, or SAIL, which gives all nine current members — a group that includes Block and Figma — a non-exclusive license to patents held by other members, as long as use is limited to AI foundation models and enabling software, including for model training.
The framework excludes licensing for hardware, user applications, user interfaces, and domain-specific implementations of foundation models.
The founding members, who make up SAIL’s board of directors, started working on the initiative about a year ago, said Linda Biel, managing director of Jamster Capital, which was hired to manage SAIL.
Jamster Capital was founded by John Amster, a co-founder of RPX Corp., a well-known defensive patent aggregator.
“This is about allowing the investment in the industry to go into innovation that benefits the public and not to legal friction,” she told MLex in an interview. “Every time you get sued, every time you get an assertion letter, you have to spend money on that. So why not just take as much of that off the table as you can so that you can direct those resources, those people, and those dollars into innovative things that help further your company and further the AI environment?”
Unlike a patent pool, there are no royalty payments. Instead, there is an annual fee of $25,000, which is a fraction of the millions typically spent resolving patent disputes.
— High stakes —
The stakes are high for companies developing foundational AI models.
Businesses around the world are increasingly incorporating AI into their daily work and investment is surging. US private AI investment grew to $109.1 billion in 2024, according to Stanford University’s 2025 AI Index Report.
In the past two years, foundational AI has moved from experimental systems to deployed business tools, driving investment, intensifying competition and prompting efforts to shape the technology’s underlying architecture.
A review of LexisNexis PatentSight+ data provided by LexisNexis Intellectual Property Solutions shows that some 95 percent of Anthropic’s patent portfolio is for AI and machine learning. About 41 percent of the portfolio would likely qualify as “covered technologies” under Section 6.5 of the SAIL membership agreement.
Although Anthropic offers the largest share of its stack among current members, Microsoft and IBM have larger patent portfolios. AI and machine learning patents represent as much as 14.01 percent of Microsoft’s and 15.44 percent of IBM’s total portfolios.
The latter two tech giants also lead in the LexisNexis Patent Asset Index, a key indicator of portfolio strength that captures not just the number of patent families but also their technological impact and strategic importance.
Somewhat less clear is the value-add of contributions by Figma, Block and Genentech, which to date have not been active in patenting AI enabling technologies. Instead, PatentSight+ data shows their portfolios are largely tied to subject matter outside SAIL’s model or disclose application-specific AI use.
Rapid advances in performance are also fueling a race toward more general-purpose AI systems, sometimes referred to as artificial general intelligence, or AGI, which is broadly defined as systems capable of learning and applying knowledge across a wide range of tasks.
Anthropic is “really focused on the pace of progress that their company is able to make,” according to Daniel Castro, vice president of the Information Technology & Innovation Foundation, or ITIF, a non-partisan policy think tank in Washington.
“They're looking at maybe whether any of the factors holding back their advancement, and right now, I would suspect that they might say, well, one of the throttle points might be the legal side of this licensing.”
— Snowball effect —
The model for SAIL is the Open Invention Network, or OIN, which was set up to protect open-source software from patent risks, Biel said. The OIN has thousands of members who agree not to assert patents against the Linux ecosystem in exchange for access to a shared patent pool and protection from other members.
The goal of SAIL should be the same as the OIN, according to Castro.
If the commons gathers momentum, he says membership could “snowball.”
“At some point, it just becomes much better for most companies to join because then they can do whatever they need in that space without having to go through separate licensing,” Castro said.
Unlike open-source software, foundational AI remains a fast-evolving and highly competitive field, raising questions about whether a similar model can take hold.
SAIL’s nine members hold more than 33,000 patents on foundational AI, but the goal is to bring more companies on board soon, according to Biel. Although declining to identify any companies by name, she said SAIL has done “a pretty thorough job of contacting” those who may be “potentially interested.”
In addition to OpenAI, other notable non-members, for now, include Alphabet (Google), Nvidia and xAI. They did not respond to an MLex request for comment.
“Several top firms want proprietary tools because those businesses are doing well and want to maintain their leadership positions,” said Darrell West, a senior fellow and AI specialist at the Brookings Institution think tank in Washington. “They feel their products have distinctive advantages and therefore are less likely to support open source or industry-wide standards.”
— Competition with China —
All the current SAIL members are North American companies and there have long been concerns about China’s stance on intellectual property in the tech sector. It is unclear whether Chinese companies such as DeepSeek or Qwen would be allowed to join SAIL.
“We certainly haven't ruled it out,” Biel said. “We want to be very, very open, but we also don't want to open a path to somebody who might have a history of not respecting IP rights.”
Competition between the US and China on AI has been fierce, and SAIL could give the American companies an edge.
“It's notable that they're not bringing in Chinese companies,” Castro said. “The US has strong IP laws, and without something like this [SAIL], they could be basically spending a lot of their time suing each other … their Chinese competitors might not be doing that.”
Yehua Zhang, an IP and tech lawyer who practices in China, explored the significance of SAIL for the Chinese market in an analysis she posted on LinkedIn.
“Faced with SAIL — a patent alliance dominated by Western tech giants and characterized by significant defensive barriers — China’s AI industry should not limit its response to passive defense,” she wrote.
Among other actions, China’s government should “establish a domestic AI foundational model patent sharing pool modelled after the SAIL framework,” she wrote.
China leads the world in the volume of AI patents, holding 74 percent of those filed from 2010 to 2024, according to Stanford’s 2025 report.
— IP clout —
While SAIL is not a patent pool and currently has just nine patent-sharing members, it does have considerable IP clout. IBM, Meta and Microsoft hold thousands of AI or machine learning patents. IBM patents will make up about 60.2 percent of those that qualify for licensing within SAIL, and Microsoft 31.7 percent, dwarfing Anthropic’s 0.3%.
However, as mentioned above, Anthropic is more exposed, with 41.2 percent of its portfolio likely to qualify for use under the criteria. About 4.3 percent of the total IBM portfolio would qualify, 3 percent of Microsoft’s, and 1.7 percent of Meta’s.
Several US companies outside of SAIL also have a considerable collection of AI and machine learning patents. Alphabet, Nvidia and Amazon are among them. Compared with global competitors, the SAIL patent collection is well-placed both on quality and number of patents, with the collection of Chinese patent owners appearing as a close competitor.
Ultimately, it would be good for the US AI industry to avoid the higher costs of proprietary software and shape the future framework, said West.
“If there is one industry standard, it would make it easier for others to build on that base.
Please email editors@mlex.com to contact the editorial staff regarding this story, or to submit the names of lawyers and advisers.