Overview
Opaque Systems turned UC Berkeley research into software for computing on sensitive data inside protected hardware. Its product evolved from shared analytics to AI workloads and evidence of agent behavior. On October 13, 2026, it announced an agreement with Anthropic. Reporting describes technology licensing and hiring part of the team; Opaque now says its operations are winding down. [1][2]
Opaque made difficult research usable through deployment software, partnerships, and verifiable records. Its technology and people continued toward a model provider as it wound down its standalone operation. Public evidence does not establish the financial return or explain why this agreement beat continued independence.
Image 1 / 1
Founding Story
The research preceded the company. Wenting Zheng and colleagues presented the Opaque encrypted analytics system at NSDI in 2017. Raluca Ada Popa dates her company co-founder role to 2021. The 2023 CEO announcement names Popa, Ion Stoica, Rishabh Poddar, Chester Leung, and Zheng as founders. Poddar studied under Popa at Berkeley after working at IBM Research. [3][4][5][6]
The original problem was concrete: organizations needed shared analysis without disclosing each participant's records. In 2022, Popa described the opportunity as “performing collaborative analytics and AI on confidential data.” Stoica said: “The world of data analytics is colliding with stricter data security requirements.” [7]
Aaron Fulkerson joined as CEO in October 2023 to commercialize that research. His later account says the technical founders wanted an operator to build the company and market. [8]
Timeline
- 2017: Berkeley researchers present Opaque at NSDI. [3]
- 2021: Popa begins her company co-founder role. [4]
- June 2022: Opaque announces a $22 million Series A. [7]
- February 2026: It announces $24 million in Series B funding at approximately $300 million post-money. [9]
- May 2026: It acquires cryptographic AI technology from Abu Dhabi's Technology Innovation Institute. [10]
- October 2026: The Anthropic agreement is announced; Opaque subsequently posts its wind-down notice. [1][2]
What They Built
Ordinary encryption protects stored files and network traffic. Confidential computing also protects data during processing through hardware-isolated environments. Opaque added software for shared analysis, access policies, and operation across protected machines. Its early platform used familiar tools such as Spark and notebooks. [7]
The 2017 research also addressed access patterns: even protected calculations can reveal information through how they read data. Its published benchmarks showed that the performance cost varied with the protection and workload. [3]
By 2026, the company described controls that checked the execution environment before a workload, enforced rules during processing, and exported evidence afterward. Its AMD collaboration illustrated these controls with sensitive debt-management files. The TII technology purchase added encrypted training and protections intended for quantum-era threats. These were vendor-described capabilities, not independent security certifications. [11][10]
Market Position
Target Customers
Opaque targeted organizations with sensitive records and restrictions on sharing them. Financial services and healthcare were early examples. Its AMD case described a credit-management company coordinating files across many partners. [7][11]
Market Size
The reviewed evidence does not establish Opaque's addressable paid market. Large estimates for confidential computing do not measure demand for its particular software or implementation work.
Competition
Cloud and chip providers supplied the underlying protected environments. That gave Opaque infrastructure to build on, while leaving platform owners able to extend their own controls. Its proposed value lay in connecting the hardware to usable workflows and evidence. The later AgenTrust specifications compose vendor attestation systems rather than replacing them. [7][12]
Business Model
Microsoft's partner documentation described Opaque Workspaces and Gateway and directed customers to Azure Marketplace. That distribution channel fits enterprise selling, but public contract prices and software-versus-services revenue were not found. The 2026 funding announcement reported $55.5 million raised. That figure measures financing, not sales or the value of the Anthropic agreement. [13][9]
Traction
Intel Capital's 2022 announcement named Ant Group, IBM, Scotiabank, and Ericsson as adopters of the technology. It did not disclose paid contract values. Fulkerson later reported adoption across government and enterprise organizations. Revenue, retention, and deployment counts remain undisclosed in the reviewed material. [7][8]
Post-Mortem
Useful infrastructure can outlast its vendor
Opaque's progression suggests a plausible structural mechanism: protecting data created demand for proof about the whole AI workflow. Hardware isolation alone could not answer which policies governed an agent or what software actually ran. Opaque expanded toward those questions before its Anthropic agreement. This interpretation fits the product sequence; it does not establish the buyer's private rationale. [14]
Commercial weakness is another possible explanation for ending independent operations. No reviewed revenue, cash-runway, or deal-price evidence resolves that alternative. A February fundraising valuation cannot establish October investor returns.
The organizational cost is clearer. Fulkerson wrote on October 13: “Not every OPAQUEling is joining Anthropic.” Customer support and migration commitments were not disclosed in the reviewed announcement. Meanwhile, Opaque's site directs open-source users to AgenTrust and says the Confidential Computing Summit continues with the Linux Foundation. [8][1]
Key Lessons
- Make research operable. Opaque connected protected computation to existing analytics tools and customer workflows. Technical capability needed a deployment path. [7]
- Define what proof covers. Its continuing TRACE work explicitly excludes model correctness and several hardware threats. A signed record supports specific claims, not a blanket safety guarantee. [12]
- Plan continuity for each constituency. Open-source projects, employees, and enterprise customers face different outcomes when a vendor winds down. Opaque identified continuing homes for community work; public customer-transition detail remains a gap. [1][8]
Sources
- OPAQUE
- Anthropic licenses, releases Opaque in acquihire
- Opaque: An Oblivious and Encrypted Distributed Analytics Platform
- Raluca Ada Popa's Homepage
- Honored to join Opaque Systems as CEO
- Rishabh Poddar
- Opaque Systems Raises $22 Million in Series A Funding To Bring Scalable, Multi-Party Analytics and AI to Confidential Computing
- I’m Joining Anthropic
- OPAQUE Raises $24M Series B at $300M Valuation to Advance Confidential AI for the Enterprise
- OPAQUE Acquires Abu Dhabi-Developed Cryptographic AI Technology from TII, Extending Confidential AI Across the Full Lifecycle with Post-Quantum Protection
- From Risk to Resilience: Confidential Computing with AMD and OPAQUE
- TRACE v0.2
- Opaque Systems, Inc.
- From Berkeley Research to the Future of AI: OPAQUE’s Next Chapter