The team behind Opaque Systems, a startup that built technology for using sensitive data in AI systems while hardware keeps it protected, is joining Anthropic, and Opaque is winding down, the company said in an Oct. 13 announcement that SecurityWeek reported Monday. The move puts expertise in protecting data during AI processing, and in proving how AI systems behave, inside a leading model developer at a time when enterprise buyers want stronger guarantees about where their data goes.
What Happened
Opaque’s website says: “Our team is joining Anthropic to carry our work forward, and OPAQUE Systems is winding down its operations.” SecurityWeek described the deal as an acquisition and said financial terms were not disclosed. Opaque’s announcement gives no terms, timeline or customer-transition details, and SecurityWeek’s report carries no comment from Anthropic.
Opaque began about a decade ago as a research project at the University of California, Berkeley’s RISELab. Its co-founders include Ion Stoica, a co-founder of Databricks and Anyscale, and Raluca Ada Popa, who now works at Google DeepMind; Aaron Fulkerson has been chief executive since October 2023. SecurityWeek reported that Opaque raised a $24 million Series B round this year at a $300 million valuation and listed Anthropic, ServiceNow and Accenture among its customers.
Opaque’s site says several of its open-source projects, including Confidential MCP, TRACE and the Weight Custody Manifest, will continue under their existing licenses through an effort called AgenTrust, and that its Confidential Computing Summit is moving to the Linux Foundation. The site does not say whether its commercial products will keep running. MCP, the Model Context Protocol, is a standard for connecting AI models to outside tools and data.
Why It Matters
Data is usually encrypted while stored and while traveling over a network, but it is typically decrypted to be processed. Confidential computing closes that gap by running work inside isolated, hardware-protected areas of a processor, so data stays shielded even from the operators of the machine. For a company sending contracts, patient records or source code to an AI model, that is the stage where exposure is hardest to control.
Opaque later moved into what it calls verifiable AI: replacing trust assumptions with cryptographic evidence of what ran, where and under which policies. That is the kind of proof auditors, regulators and large customers may ask AI providers to produce.
For Opaque’s customers, the immediate issue is continuity. Neither company has said what the wind-down means for existing deployments, so organizations that rely on its software will need answers on support, data handling and migration directly from the company.
What to Watch
- Whether Anthropic confirms the deal and adds attestation or confidential-computing features to its products.
- What Opaque tells customers about support and migration.
- Whether the open-source projects keep active maintainers after the transition.
