Skip to content
be-human.ai
← Back to the journal

Human oversight

Board Charter for Company Superintelligence: Roles and Review Dates

A short board charter for company-level AI capability: who owns it, who approves changes, and how often the board reviews it. No timeline predictions.

Be Human typographic cover with the title: Board Charter for Company Superintelligence: Roles and Review Dates
Conceptual illustration created with AI · Be Human

A board charter for company-level AI capability is a short written document that names who owns the capability, who may approve changes to what an agent can see and do, and how often the board reviews both. Write it before any agent receives wider access, and base it on the capability that exists today rather than on forecasts about when stronger systems will arrive. A few pages, signed by a board sponsor and tied to fixed review dates, is enough for most small and mid-sized companies to start governing AI responsibly.

Key takeaways

  • Name one accountable board sponsor and one operations owner for the company's AI capability, each with a named backup.
  • Separate the authority to approve new tools, data sources and memory scopes from the authority to run routine agent work.
  • Build every review around evidence: logs, evaluation results, incidents and permission changes.
  • Set review dates by calendar and by trigger, so an incident or vendor change forces a review between scheduled meetings.
  • Write the charter around observable controls, not predicted timelines.

What a board charter must cover

Most companies do not need a governance manual. They need a document that answers five plain questions: what the AI capability is for, who answers for it, what it is allowed to touch, who can change those limits, and how the board will know whether the limits still hold. The NIST AI Risk Management Framework groups AI risk work into four functions: govern, map, measure and manage. The govern function assigns responsibility for the other three, and a board charter is the short form of that assignment. The OECD AI Principles also name accountability as a core value, which means a person or body must be identifiable for the outcomes an AI system produces.

The charter should stay short enough that a board member can read it in one sitting. If it grows beyond that, the detail usually belongs in an operating procedure that the charter points to. For a broader view of how responsibility is split across the company, AI Governance Board: How a Small Company Approves Agent Upgrades shows one practical arrangement.

Five roles the charter names

The charter names roles rather than job descriptions. Each role needs a named holder and a backup who has actually practiced the work.

  • Board sponsor: holds final accountability, receives the review summary and decides disputed approvals.
  • Capability owner: decides what the agent is for and accepts its business outcomes and customer-facing commitments.
  • Operations lead: keeps the agent running, reads its logs, runs evaluations and opens incident records.
  • Data and privacy steward: confirms which personal and company data the agent holds, where its memory is stored and how long it is kept.
  • Security reviewer: checks tool credentials, permission scope and exposure to outside content such as incoming email or web pages.

Small companies often combine roles, and that is workable if two rules survive the combination. The person who approves a change should not be the only person who reviews its logs, and no backup should exist only on paper. For a sharper view of where authority should sit, Control Points for Superintelligent Agents: Five Places to Hold Authority is a useful companion.

Approval authority: who signs for what

Approval authority should follow the risk of the change, not the seniority of the person asking. The guide below can sit inside the charter as a decision table, written here as a list.

  • Changes to the agent's purpose, its customer-facing commitments or its audience go to the capability owner and the board sponsor.
  • New tools, new data sources or wider memory scopes go to the data steward and the security reviewer before activation.
  • Permissions that allow payments, outbound messages to customers or deletion of records require two-person approval.
  • Routine changes to wording, report formats and schedules go to the operations lead and are logged in the change record.
  • Any matter involving a legal hold, a regulated category of personal data or a formal complaint goes to the board sponsor the same day.

The OWASP authorization guidance points in the same direction: grant the minimum access a task needs, and deny anything not explicitly allowed. Apply that rule to the agent the way you would apply it to a new employee. Thresholds for the two-person rule deserve their own review, and Two-Person Approval for AI Agents: Which Actions Need It covers them in detail.

Review cadence: when the board looks

A review date that exists only as a sentence in a document tends not to happen. Put each review in the company calendar with an owner and an agenda, and keep three kinds of review.

  • Monthly: the operations lead sends a written summary covering incidents, permission changes, evaluation results and open exceptions.
  • Quarterly during the first year, then twice a year once operations are stable: the board reviews the summary, confirms role holders and decides whether any permission group should change.
  • Event-driven: an incident, a change in a model or tool provider, a new data source, a customer complaint, or the departure of a staff member who had tool access triggers a review within the same week.

Evaluations feed the monthly summary. Anthropic's guidance on evals for AI agents describes measuring agent behavior with defined tasks and repeatable checks rather than relying on a few impressive demonstrations. The operations lead also needs records to read, and the OpenTelemetry observability primer explains the logs, traces and metrics that make such records useful.

Nothing in this cadence depends on predicting when more capable systems will appear. It depends on what the company has observed in its own operation, which is the only evidence a board can check.

Step by step: adopt the charter in one meeting

A first charter can be adopted in a single meeting if the material is prepared beforehand.

  • Write the purpose of the capability in two sentences, and state what the agent must never do.
  • Name the five roles, with a backup for each, and confirm in writing that each person accepts the role.
  • List every tool, data source and memory scope the agent holds today, along with the credential each one uses.
  • Sort each permission into one of three groups: runs without approval, runs after approval, or not permitted.
  • Apply the approval guide to each group and record who signs for each change type.
  • Enter the monthly, quarterly and event-driven review dates in the company calendar with a named owner.
  • Define the evidence pack each review must include: a sample of logs, evaluation results, the incident list and the permission changes.
  • Have the board sponsor sign the charter, then store it where the next board member can find it.

Illustrative scenario: a support agent asks for more access

This scenario is illustrative and does not describe a specific company. A customer support agent drafts replies and reads order status. The operations lead proposes read access to the billing system so the agent can answer refund questions. Under the charter, this is a new data source. The data steward checks which personal data the billing records expose, and the security reviewer checks the scope of the credential. The request waits for both approvals, and the board sponsor sees it in the monthly summary.

Shortly after the change is approved, an evaluation shows the agent quoting a refund rule that changed the previous month. The event trigger forces a review before the next scheduled meeting. The team keeps the agent on order status only, corrects the rule source, records the incident and reruns the evaluation before any refund access is reconsidered. The charter did not prevent the error, but it made the error visible, assigned it to a named owner and limited its reach.

How Be Human's managed approach handles this

Be Human designs and operates personalized AI agents whose identity, memory, channels, tools, permissions and operating context stay under human control. A first engagement defines one useful role, the memory and tools it needs, its approval boundaries and an observable success criterion, which gives a charter something concrete to govern. Managed operations can include monitoring, recurring evaluations, integration maintenance, permission review, incident response and controlled improvements, so the operations lead and the security reviewer work from a standing calendar rather than a document read once. Be Human does not set the company's risk appetite; the board does, and the charter records that decision. If you are unsure which task should go to an agent first, take the free Be Human AI diagnostic to find the first task worth handing to an agent.

Frequently asked questions

Does a small company really need a board charter for AI? Yes, if AI tools can read company data, send messages or change records. The charter does not need to be long, but it should name who owns the capability, who approves changes and when the board reviews the evidence. Without that, decisions are made by whoever installed the tool last.

Who should sign the charter? The board sponsor should sign it, because the sponsor carries final accountability and resolves disputes. The capability owner and the operations lead should also acknowledge their roles in writing. The named roles, backups and calendar entries matter more than the signatures themselves.

How does a charter differ from an AI use policy? An AI use policy tells employees what they may do with AI tools in their daily work. A board charter governs the company's own agent capability, including its purpose, permissions, approvals and reviews. Many companies need both, and the use policy can refer to the charter for agent-specific rules.

Should the charter mention future superintelligent systems? It should not depend on predictions about them. Write the charter around the capability you run today and the controls you can observe. When the capability changes, the event-driven trigger forces a review, and the charter is updated with evidence rather than with assumptions about timing.

Your next step

See where an AI agent fits in your company first.

Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →