Authority before commitment™

Decide which changes a system may make, and under what authority, before they happen.

That is constitutional computing™. Its first product, Inktomi, applies it to enterprise AI agents: each consequential action an agent proposes, such as a financial transfer, a regulatory filing or a patient record change, is checked against your policy before it runs. Every decision is recorded.

Problem

The instant before the world changes.

Commitment is the instant a proposal becomes a fact in the world: the write lands, the payment clears, the record is amended. AI agents now make those changes using credentials granted weeks ago. Those credentials establish access. They do not, by themselves, establish authority for that exact change, at that moment.

Valid credentials do not make an action right.

An agent with access to your payment system can pay any invoice it can reach. Access was settled when it was set up. Whether this payment, to this vendor, should happen now is a separate question, and a valid credential does not answer it.

  • Pay invoice 4471: $48,200 to Northwind SupplyCredentials valid
  • Submit the quarterly filing to the regulatorCredentials valid
  • Change the medication list in a patient recordCredentials valid

No check at the moment of action

Many AI systems grant permission once and assume it indefinitely.

Changes that go around the check

An agent edits a file with its own shell or editor, so no check sees the change before it lands. How Inktomi finds them afterwards

Evidence assembled by hand

When an auditor asks who approved an action, the answer is spread across logs and inboxes, sometimes taking weeks to interpret.

Regulatory€35Mor 7% of global turnover

The maximum fine for prohibited AI practices under Article 5 of the EU AI Act.

EU AI Act, Article 99
Security+$670Kper breach

The added cost of a breach at organizations with high levels of unsanctioned AI use.

IBM, Cost of a Data Breach 2025
Governance63%lacked an established AI governance policy

Among breached organizations studied, 63% either had no AI governance policy or were still developing one.

IBM, Cost of a Data Breach 2025

Why now

Four frontier labs disclosed incidents in ten weeks.

Most began inside a boundary that was meant to hold, usually a test environment. It failed. What the agent did next was decided by the agent.

“Since August, public fears have grown over the potentially serious, even life-threatening, impacts of AI tools falling outside of human control.”

BBC News, 25 September 2026 · Read the report
21 JulOpenAIModels escaped an isolated test environment and compromised a third party's production infrastructure.
30 JulAnthropicClaude models reached the production systems of three organisations during cybersecurity evaluations.
5 AugMetaMuse Spark 1.1 reached the internet and exploited a third-party service during a cybersecurity test.
18 SepGoogleGemini reached three outside systems by guessing credentials and using passwords found in public.

All seven incidents, with sources, and where Inktomi would have applied

Inktomi, the first product

Policy decides the routine cases. People decide the rest.

Inktomi is the first product built on constitutional computing. It sits between what an agent proposes and what your systems carry out. Your identity and approval controls stay in place.

  1. Step 1The agent proposes an action

    A specific call with specific details: this payment, this amount, this vendor.

    This payment, this amount, this vendor.

  2. Step 2Inktomi checks it against policy

    Routine cases are approved automatically. Exceptions go to a human approver.

    Routine cases pass. Exceptions go to a person.

  3. Step 3Your system acts only with a permit

    Every connected system verifies the permit itself, so protection holds at every connected enforcement point.

    A call that goes around Inktomi is refused.

One payment, three outcomes

The same request from the finance agent: pay invoice 4471, $48,200 to Northwind Supply. What changes is the situation around it.

Permitted

Within the agent's $50,000 limit. Vendor verified.

Policy approves it on the spot. The connected payment system checks the permit and pays. No one had to click anything.

Held for approval

New vendor, over $25,000.

Policy requires the controller's sign-off. The approval is tied to this payment only and cannot be reused for another.

Refused

Vendor bank details changed seconds ago.

Policy blocks payment until the change is verified. With no permit, the connected payment system refuses the call.

Your governed decisions, in one place.

The console displays what each agent asked to do, what was decided and why, and what is waiting for a person. Each decision is kept in a tamper-evident record an auditor can replay.

Take a closer look at the console

The Inktomi console listing recent agent decisions: permitted, held for review and refused, with the reason for each. Enlarge

Kihan™'s ambition

A foundational layer for authorized computation.

Enterprise AI is the first application. Constitutional infrastructure is the ambition. Kihan is building constitutional computing infrastructure; Inktomi is its first product.

How constitutional computing works · How Inktomi applies it

Deployment

What your team installs

Inktomi runs on your infrastructure and connects to the systems you already use.

  • One lightweight node on each agent host. It checks actions before they run.
  • Your identity provider signs in approvers, over OIDC.
  • Your SIEM receives every decision as a standard event, in OCSF.
  • A coverage report shows which paths are governed and which are not.

Where each piece runsSee integrations and status

Integrations
Connects toExamplesStatus
Agent hosts[Claude Code, Codex, …]Available
Tool servers (MCP)Your MCP serversAvailable
Identity providersAny OIDC providerAvailable
SIEMOCSF compliantAvailable
Payment, records and filing systemsYour systems of recordCustomer integration

Industries

Built for consequential, regulated workflows.

Built by a Silicon Valley team with experience at companies like Apple, Cisco & Sony. A two-part article describing the architecture has been accepted for publication in IEEE Computer, December 2026 and January 2027.

A Silicon Valley team

Our team has built solutions that underpin some of the largest enterprises and internet service providers in the world, with deep expertise and experience gained at companies like Apple, Cisco & Sony.

About Kihan

Forthcoming in IEEE Computer

A two-part feature by Kihan's founder, Jonathan Hirshon, on the architecture behind Inktomi has been accepted for publication by IEEE Computer, appearing in December 2026 and January 2027.

Read about the research

Request a demo.

Request a 30-minute live demonstration with Kihan, with time for your questions.