Kihan Inktomi Review: Enforceable Authority for Enterprise AI Agents

As artificial intelligence agents move from answering questions to executing transactions and modifying production systems, a critical security gap has emerged: the difference between an AI agent’s capability to perform an action and its authority to do so. Kihan, a new San Francisco company, has introduced Constitutional Computing, a mathematically based architecture designed to make AI governance enforceable at the point of execution. Their first product, Inktomi, aims to prevent unauthorized AI agent actions before they can affect enterprise systems, addressing a problem that research shows has already impacted nearly half of organizations.

Kihan’s approach is fundamentally different from conventional AI guardrails, which typically rely on instructions asking the model to behave. Instead, Constitutional Computing establishes an independent authorization system outside the AI itself, where authority is written down, signed, and enforced separately. This architecture ensures that each consequential action is evaluated against governing authority before execution, with a cryptographically verifiable permit bound to the specific operation and its parameters. The result is a system where no valid permit means no execution on protected paths.

What is Kihan Inktomi?

Inktomi is Kihan’s first product, bringing Constitutional Computing to enterprise AI environments. It is designed to work with agents built on models from Anthropic, OpenAI, and Google, as well as agent frameworks like Hermes Agent and tool ecosystems using the Model Context Protocol (MCP). The product’s decision engine, already operational, evaluates specific actions against signed rules, determines whether to permit, refuse, or escalate a request, and preserves a verifiable decision record.

Production enforcement integrations are under development, with initial enterprise evaluations targeted for December 2026. Inktomi is designed to run on customers’ own infrastructure, complementing existing identity management, cybersecurity, and AI governance technologies rather than replacing them. This makes it a practical addition for organizations already investing in security and compliance frameworks.

What Stands Out

Independent Authorization Outside the AI

Unlike guardrails that ask the model to comply, Inktomi’s rules sit outside the model. The AI never holds the authority, so even if the model is persuaded, confused, or wrong, the authorization remains intact. This separation of capability from permission is the core innovation.

Cryptographically Verifiable Permits

Each permitted action receives a signed single-use permit that the executing tool verifies before acting. This ensures that only explicitly authorized operations proceed, with tamper-evident records showing what was authorized, refused, or referred for human approval.

Protection Against Direct Calls

Unlike gateways that only control traffic routed through them, Inktomi requires a permit at the tool level. A tool that requires an Inktomi permit refuses a direct call that bypasses the gateway, closing a common security loophole.

Designed for Regulated Industries

Kihan is initially targeting financial services, insurance, healthcare, legal services, and other regulated industries where compliance and auditability are paramount. The architecture aligns with emerging regulations like the EU AI Act, which requires human oversight and record-keeping for high-risk systems.

Who It’s For

Inktomi is designed for enterprise organizations deploying AI agents in production environments where unauthorized actions could have serious consequences. This includes companies in financial services handling transactions, healthcare organizations managing sensitive patient data, legal firms dealing with confidential information, and any regulated industry where AI governance is becoming a compliance requirement.

It is also suited for security and compliance teams who need to enforce policies consistently across multiple AI models and frameworks. The product complements existing governance tools, making it a natural addition for organizations already using platforms like IBM watsonx.governance or NVIDIA OpenShell.

Pricing and Availability

Kihan has not yet announced pricing for Inktomi. The product is under development, with initial enterprise evaluations targeted for December 2026. Organizations interested in early access can visit Kihan’s website to arrange a live demonstration and learn more about the architecture.

What Makes It Worth Considering

Kihan’s Constitutional Computing represents a fundamental shift in how AI governance is approached. By separating authority from capability and enforcing it at the execution point, the architecture addresses a problem that traditional guardrails and gateways cannot solve. The research underlying the architecture has been accepted for publication in IEEE Computer, the flagship publication of the IEEE Computer Society, lending academic credibility to the approach.

The product is designed to run on customer infrastructure, ensuring data sovereignty and alignment with existing security policies. Its ability to integrate with multiple AI models and frameworks makes it a versatile solution for heterogeneous enterprise environments. For organizations facing the growing challenge of AI agent governance, Inktomi offers a mathematically rigorous and practically enforceable approach.

Verdict

Kihan Inktomi offers a groundbreaking, mathematically rigorous solution for enforcing AI agent permissions, ideal for regulated industries. By separating capability from authority and providing cryptographically verifiable permits, it addresses a critical security gap that traditional guardrails cannot close. While still in development, the architecture’s academic foundation and practical design make it a promising tool for enterprise AI governance.