Opaque Systems announced a new agreement with Anthropic to transfer part of its business to the AI company in a deal combining the licensing of Opaque technologies with the recruitment of some of its employees, rather than a traditional full acquisition of the company. The list includes CEO Aaron Fulkerson and CTO Rishabh Poddar, according to The Information.
Opaque described the agreement as a step that “moves its work forward,” without disclosing financial details or clarifying the precise scope of the assets, teams, and software covered. The arrangement gives Anthropic access to specialized expertise in confidential AI, a field aimed at running models and workloads on sensitive data while keeping the data protected even during use.
What is confidential AI?
This approach relies on confidential computing capabilities to protect data within hardware during processing, not only while it is stored or transmitted. According to Arjun Jaggi, associate director of research and development innovation at Cognizant, confidential computing enables workloads to run on sensitive data while the data remains protected by the hardware itself during use.
The technology’s importance is linked to AI applications that need to analyze highly sensitive information while reducing the trust required in the infrastructure layers and the entities managing them. However, the current agreement does not yet explain how Anthropic will integrate Opaque’s technologies into its products or services.
What was Opaque building?
Opaque began as a research project at the RISELab at the University of California, Berkeley, about 10 years ago, then developed work around verifiability in the AI stack and provable governance. Earlier this year, the company closed a $24 million Series B funding round at a valuation of $300 million.
Opaque also founded the Confidential Computing Summit, whose management it said had been transferred to the Linux Foundation, which co-hosted the June edition of the conference. In collaboration with the Technology Innovation Institute in the UAE, the company worked on an open standard for verifiable governance in AI.
In June, Opaque launched the open-source AgenTrust project for verifiable agent governance, in addition to Confidential MCP, which it describes as a secure and verifiable implementation of the MCP protocol.
Why does the deal matter to developers?
The agreement is significant because it connects specialized expertise in protecting data during processing with one of the largest AI labs. This could accelerate the development of more verifiable architectures for running models and agents on sensitive data.
But the practical impact on developers depends on the fate of Opaque’s open-source projects. The source does not specify whether these projects will remain independent, be transferred to Anthropic, or have their licenses changed, nor does it clarify how many employees will join the company. The continued existence of AgenTrust, Confidential MCP, and the ecosystem around them therefore remains an open question.
Anthropic’s previous interest
The agreement did not come out of nowhere; in 2024, Jason Clinton, Anthropic’s chief information security officer, spoke on Opaque’s AI Confidential podcast about the importance of confidential computing to the future of AI. Nevertheless, the available information does not reveal whether the deal will lead to an announced product or a specific timeline for integrating Opaque’s technology.