Anthropic Acquires Confidential AI Startup Opaque Systems

AI safety and research company Anthropic has acquired Opaque Systems, a startup known for developing technology that enables artificial intelligence to process sensitive data without exposing the underlying information. The move is expected to bolster Anthropic’s capabilities in securing AI deployments and verifying the behaviour of autonomous agents.

Opaque Systems, which originated as a research project at UC Berkeley’s RISELab, has spent the last decade focusing on confidential computing. The company’s technology allows AI systems to perform computations on protected data while keeping it encrypted, often leveraging specialized hardware to maintain security. Following its acquisition, Opaque Systems has announced that its team is joining Anthropic and that its commercial operations are winding down.

The integration comes as Opaque Systems has increasingly shifted its focus toward verifiable AI. This approach seeks to replace traditional trust assumptions with cryptographic evidence, allowing organizations to independently audit exactly what code an AI system ran, where it executed, and whether it adhered to established security policies. This transition reflects a broader industry push to move beyond simple data protection toward ensuring the integrity and transparency of AI agents.

Before the acquisition, Opaque Systems had been actively developing open-source tools to address trust and security gaps in the AI ecosystem, including the Agent Governance Toolkit and various projects related to confidential computing, such as TRACE and the Confidential MCP. It remains unclear how these community-focused projects will be managed under Anthropic’s stewardship.

The startup, which recently raised a $24 million Series B funding round, counts large enterprise organizations among its former customers. While the financial terms of the transaction were not disclosed, the move underscores the growing strategic importance of confidential computing in the enterprise AI sector, where companies are increasingly concerned about the privacy and security risks associated with sharing sensitive data with large language models.