Anthropic absorbs Opaque, a confidential AI startup

Opaque Systems, a startup that lets companies run AI on sensitive data without exposing it, is winding down as its team joins Anthropic. "Our team is joining Anthropic to carry our work forward, and OPAQUE Systems is winding down its operations," the company says on its website. Neither side has disclosed the terms.

The deal brings Anthropic a team that built ways to prove what code ran on a customer's data. Constellation Research ties it to Anthropic's push toward large customers that want AI on their own premises and under their own control.

Acquisition or acquihire

Reports describe the deal in two ways. SecurityWeek, reporting on October 19, 2026, calls it an acquisition. Constellation Research reported on October 13 that Anthropic is licensing Opaque's technology and bringing on some of its team. The New Stack also called it an acquihire, a deal where a company hires a startup's staff rather than buying the business outright.

Both accounts agree on the outcome. Opaque's people move to Anthropic, and Opaque Systems itself shuts down. SecurityWeek says Opaque announced the deal the week before its report.

The structure resembles other recent deals that license a startup's technology and hire its people, such as Apple's agreement with Huxe.

What Opaque built

Opaque started about ten years ago as a research project in UC Berkeley's RISELab. Its founders include Ion Stoica, a co-founder of Databricks and Anyscale, and Raluca Ada Popa, who now works at Google DeepMind. Aaron Fulkerson became CEO in October 2023.

The company worked on "confidential computing." In plain terms, the data stays encrypted and walled off by the processor itself, even while a program uses it. Opaque applied that to AI, so sensitive data could feed an AI system while the hardware kept it protected. Constellation says Opaque worked with Nvidia, Intel, Microsoft and Google on the technology.

Later, Opaque moved toward what it calls "verifiable AI." Its stated aim was "replacing trust assumptions with cryptographic evidence of what ran, where it ran, and under what policies." Opaque listed Anthropic among its own customers, alongside ServiceNow and Accenture, SecurityWeek reports.

Detail	What the sources say
Last funding	$24 million Series B this year
Valuation	$300 million
Origin	UC Berkeley RISELab, about ten years ago
Deal terms	Not disclosed
Company status	Winding down

What happens to its open-source work

Opaque's open-source projects do not disappear. According to the company's website, they continue at AgenTrust under their existing licenses. The list includes Confidential MCP, Confidential A2A, TRACE, Agent Manifest and Weight Custody Manifest. MCP, the Model Context Protocol, is a common way to connect AI agents to tools and data. The company says the Confidential Computing Summit continues at the Linux Foundation.

"We built a company from university research. We helped establish confidential AI as a category," Fulkerson wrote, as quoted by Constellation.

What this means for developers

For teams that use Opaque's products, or that are weighing Anthropic for sensitive data:

• Opaque customers need an exit plan now. No source says what the wind-down means for existing customers. Ask Opaque in writing how long service runs and how you get your data and configurations back.
• Check which open-source pieces you depend on. If you use Confidential MCP or TRACE, find the AgenTrust repository and confirm it is maintained before you upgrade.
• Read the deal as a signal about Anthropic's enterprise plans. Constellation links the move to customers who want on-premises deployments and "sovereign AI," meaning AI that runs under a country's or company's own control. If that is your requirement, ask Anthropic's sales team what it plans to offer and when.
• Expect proof, not promises, to become the bar. Opaque's pitch was evidence of what code ran on your data. Buyers of AI services can ask for that kind of evidence in their own security reviews.