Anthropic Acquires Confidential AI Startup Opaque Systems

Anthropic Absorbs Confidential Computing Pioneer Opaque Systems

Anthropic has acquired Opaque Systems, a decade-old confidential computing startup that built hardware-backed data protection and cryptographic verification technology for AI workloads, in a deal announced the week of October 19, 2026. Financial terms were not disclosed, and the two companies have given conflicting descriptions of the deal's structure — some reporting frames it as a straightforward acquisition, while others describe an acquihire-and-license arrangement in which Anthropic hired Opaque's CEO Aaron Fulkerson and CTO Rishabh Poddar and licensed the company's technology, while most of Opaque's remaining staff were reportedly not retained. Opaque's own announcement states its team "is joining Anthropic to carry our work forward," and that Opaque Systems is winding down its operations.

What Opaque Systems Built

Opaque originated roughly a decade ago as a research project inside UC Berkeley's RISELab, co-founded by Ion Stoica (also a co-founder of Databricks and Anyscale) and Raluca Ada Popa, an encryption and systems-security researcher who has since moved to Google DeepMind. Aaron Fulkerson took over as CEO in October 2023 and steered the company from a research spinout toward a commercial confidential-computing platform.

The core technology runs AI inference and training jobs inside hardware-sealed Trusted Execution Environments (TEEs) — isolated compute enclaves that keep data encrypted while it is actually being used, not just at rest or in transit. That closes a gap standard encryption cannot: a model has to decrypt its input to read it, and until now that decrypted data was exposed to the host operating system, hypervisor, or cloud provider during processing. Opaque's platform produces cryptographic attestation — verifiable evidence of exactly which code ran, on which machine, and under which policy — so a bank, hospital, or government agency can confirm how its sensitive data was handled without taking the AI vendor's word for it.

Opaque described this evolution as a move from "confidential AI" to "verifiable AI": "replacing trust assumptions with cryptographic evidence of what ran, where it ran, and under what policies." The company contributed several projects to the open-source ecosystem, including the Agent Governance Toolkit, the TRACE AI runtime attestation standard, Confidential MCP (released under the MIT license), and the Weight Custody Manifest framework for proving model-weight provenance. Opaque had also handed stewardship of its standards summit to the Linux Foundation.

A Deal Built on an Existing Relationship

Anthropic was not a stranger to Opaque's technology — it was already listed as an Opaque customer, alongside enterprise names like ServiceNow and Accenture. In June 2025, Anthropic published a confidential-inference design in partnership with Pattern Labs, noting that some AI accelerator chips still lacked full confidential-computing support — a gap Opaque's hardware-attestation approach was built to work around.

Opaque had also built relationships with major infrastructure providers — Nvidia, Intel, Microsoft, and Google — and had expanded its security framework into Europe and the UAE, positioning itself for the sovereign-AI and on-premise deployment market. Earlier in 2026, the company raised a $24 million Series B round at a $300 million valuation, underscoring investor confidence in confidential AI as a category shortly before the acquisition.

Why Anthropic Is Bringing the Capability In-House

As large language model providers push into regulated industries — finance, healthcare, government, and sovereign-AI deployments — they increasingly need to prove, not just promise, that sensitive customer data stays protected while a model is actively using it. Anthropic's bet is that enterprises will pay a premium for AI they can cryptographically verify rather than merely trust on a vendor's assurance. Bringing confidential-computing and attestation expertise in-house — rather than relying on an independent third party — lets Anthropic build those guarantees directly into Claude's enterprise inference stack instead of pointing customers to an outside vendor.

Impact Assessment

Recommendations

For Enterprise Security and Compliance Teams

• Do not assume Opaque's existing support contracts, SLAs, or roadmaps carry over automatically — get written confirmation from Anthropic on continuity.
• Re-examine any compliance or audit program that relied on Opaque as an independent verifier of AI vendor behavior; that independence assumption no longer holds in the same form.
• Ask any vendor making a "confidential AI" claim to specify which TEE hardware backs it, rather than accepting the label at face value.

For Current Opaque Customers

• Contact Anthropic and former Opaque leadership directly for a documented migration or support path before any existing contract lapses.
• Confirm whether self-hosting or forking Confidential MCP (MIT-licensed) is a viable fallback if commercial support disappears.
• Audit any production system built on Opaque's SDKs or APIs for a deprecation timeline.

For Claude/Anthropic Enterprise Customers

• Watch for a new Anthropic enterprise tier or feature announcement built around confidential inference and hardware attestation, and ask the account team for a roadmap and timeline.
• Factor attestation capability into vendor comparisons for workloads with regulated or sovereign data-residency requirements.

For the AI Security Community

• Track how TRACE and the Agent Governance Toolkit standards are governed now that a model vendor has absorbed their original steward; watch the Linux Foundation's role in the Opaque Summit for signs of continued independence.
• Expect more confidential-AI acquihires as model providers race to internalize trust and verification infrastructure rather than depend on third parties.

Key Takeaways

1. Anthropic acquired — or, per some reporting, acquihired and licensed the technology of — confidential computing startup Opaque Systems, announced the week of October 19, 2026, with financial terms undisclosed.
2. Opaque's CEO Aaron Fulkerson and CTO Rishabh Poddar are joining Anthropic; reports differ on whether the rest of Opaque's team is included, and Opaque Systems is winding down operations.
3. Opaque's technology runs AI workloads inside hardware-based Trusted Execution Environments, producing cryptographic attestation that proves what code processed sensitive data, where, and under what policy — a shift the company described as moving from "confidential AI" to "verifiable AI."
4. Opaque was already an Anthropic customer and had raised a $24 million Series B at a $300 million valuation earlier in 2026, with infrastructure partners including Nvidia, Intel, Microsoft, and Google.
5. The deal raises a governance question: attestation tooling is valuable because it lets customers independently verify an AI vendor, and that independence is harder to maintain once a model vendor controls the tooling.
6. Confidential MCP's MIT license should allow community continuity regardless of Opaque's corporate fate, but the future of TRACE, the Agent Governance Toolkit, and the Weight Custody Manifest standards remains unclear.

Sources

• Anthropic Acquires Confidential AI Startup Opaque Systems — SecurityWeek
• Anthropic licenses, releases Opaque in acquihire — Constellation Research
• Anthropic buys confidential AI startup Opaque to court enterprises — Dealroom.co