Opaque Systems winds down as its confidential AI team joins Anthropic

Opaque Systems winds down as its confidential AI team joins Anthropic

13 October 2026: Opaque Systems, the confidential AI company that grew out of UC Berkeley's RISELab, says its team is joining Anthropic and that the company is winding down its operations. Organisations that run sensitive AI workloads on Opaque's platform now have a vendor that is closing, with no published end date for the service.

Not yet independently verified. Anthropic has not published a statement, and the technology licence and the names of the staff moving come from press reports citing The Information. Opaque gives no date by which its service stops. We will update this when it can be confirmed, and remove this note.

What Opaque announced

The announcement went up on Opaque's site on 13 October under the title "OPAQUE's Next Chapter". "Today, we're proud to share that our team is joining Anthropic to carry OPAQUE's work forward," the company wrote. Its home page has since been replaced with a farewell that states plainly that "OPAQUE Systems is winding down its operations", followed by an archived copy of the old site, marked as preserved as it appeared in October 2026.

Neither page describes a purchase of the company, a price or any terms. Taiwan's iThome, citing The New Stack's account of a report by The Information, says Anthropic has taken a licence to Opaque's technology and hired part of its staff, including chief executive Aaron Fulkerson and chief technology officer Rishabh Poddar, and that the value and full terms have not been disclosed. That is the shape of a licence and hiring deal rather than an acquisition of the company, which is also why the business itself is closing rather than continuing under a new owner.

Ten years from research project to verifiable AI

Opaque began about ten years ago as a RISELab research project on letting organisations use sensitive data while keeping it encrypted even during processing. It turned that into a product for running AI workloads inside hardware-enforced trusted execution environments, a category it calls confidential AI. As AI agents spread, it moved on to what it calls verifiable AI: cryptographic evidence of which code ran, where it ran and under what policies, so a customer does not have to take a provider's word for how its data was handled.

The open-source side of that work does not close with the company. Opaque says the Weight Custody Manifest, Agent Manifest, Confidential MCP, Confidential A2A and TRACE "remain open and available under their existing licenses", with development continuing at AgenTrust, a community it helped start. The Confidential Computing Summit, which Opaque founded, continues under the Linux Foundation.

Where this leaves Opaque customers

Opaque's customers were mostly organisations that could not otherwise put AI near regulated data, which makes an unplanned exit harder than for an ordinary SaaS tool. Three things are worth settling with the company while its staff are still reachable:

• the date the hosted service and support end, which neither page states;
• how data, keys and attestation records held in the platform are exported or destroyed, with written confirmation of destruction for regulated records;
• whether any workloads depend on Opaque-specific components rather than on the open projects that AgenTrust now maintains.

Anthropic has said nothing about offering Opaque's technology as a product, so customers should not plan around a successor service. The move does add confidential computing know-how to a model provider at a time when enterprise buyers are asking for proof of how prompts and data are protected during inference.

Alternatives we list

Our catalogue lists Fortanix, a data security platform for encryption and key management, and Decentriq, whose data clean rooms are built for privacy-safe collaboration on sensitive data. Anthropic has its own profile.

Sources

• OPAQUE: From Berkeley Research to the Future of AI, OPAQUE's Next Chapter, 13 October 2026
• OPAQUE home page, wind-down notice (read 2026-10-22), 22 October 2026
• iThome: Anthropic取得Opaque機密AI技術與人才, 22 October 2026