Perspective: AI Operational Guidance

When you let software make decisions and act on them, you have not automated a task. You have delegated authority. Most organizations have not yet written down to whom.

Aaron Parks  ·  President, ParksPacific Financial  ·  July 2026

Clarity today. Confidence tomorrow.

For years, enterprise AI created things but had no authority to use them.  It was a person who reviewed the draft, and a person who remained the decision maker. That era is closing. Agentic AI — software that plans and executes multi-step actions across your systems in pursuit of a goal, can operate without a human approving each step.  Gartner predicts as many as 40% of enterprise applications will incorporate task-specific AI agents by the end of 2026¹, and Deloitte’s 2026 State of AI in the Enterprise survey found 74% of organizations expect at least moderate use of AI agents by 2027².

Here is the number that should concern every board: in that same Deloitte survey, only 21% of enterprises report having mature governance in place to manage agentic AI risk. Roughly 80% lack basics such as clear boundaries defining which decisions agents may make independently, real-time monitoring of agent behavior, and audit trails capturing the full chain of agent actions².

SECTION ONE

The Risk You Invited In

An agent is different from a chatbot in one decisive way: a chatbot suggests, an agent does. Modern agent platforms let you grant an agent credentials, connect it to tools — your ERP, your codebase, your AP & AR — and configure how much autonomy it has. Most platforms ship with approval gates: the agent proposes an action, a human confirms it.

Then reality arrives. Approval prompts are friction, and friction is the enemy of productivity.  And so, some teams flip the switch — “always approve,” “always allow,” full tool access — and a workflow that was supposed to assist a human quietly becomes a workflow that acts in place of one, including, sometimes, in production.

DOCUMENTED INCIDENT

When “always allow” met production

In July 2025, an AI coding agent on the Replit platform deleted a live production database containing records on roughly 1,200 executives and nearly 1,200 companies — during an explicitly declared code freeze, against repeated instructions not to make changes. The agent then misrepresented what it had done and incorrectly claimed rollback was impossible, delaying recovery³. Replit’s CEO publicly called the failure “unacceptable” and implemented emergency safeguards, including automatic separation of development and production databases⁴.

Note what this incident was not: it was not a hacker, and it was not malware. It was an autonomous system with production access, broad permissions, and no enforced human gate.

This is the anatomy of the new risk. When an agent changes production, the impact on your financials, output, or operations may be immediate — or it may surface weeks later in a reconciliation, a customer complaint, a regulator’s letter. Either way, the damage may not be reversible. And in most organizations today, three things are missing when it happens: a defined response process, a named leader who owns the outcome, and a notification path that reliably reaches the C-suite. Research from Writer found that 35% of organizations admit they could not shut down a rogue AI agent if one emerged, and 36% have no formal plan for deploying agents at all⁵.

You don’t discover your governance gaps when you deploy the agent. You discover them during the incident.

SECTION TWO

Does the Board Actually Know?

Awareness at the top is improving, but it is not yet oversight. Protiviti and BoardProspects’ 2026 global survey of 772 directors and executives found only 26% of boards discuss AI at every meeting — though among organizations reporting high AI returns, 63% do, versus 13% of low-return organizations⁶. Deloitte’s Global Boardroom Program survey found 31% of respondents say AI is not on the board agenda at all, and two-thirds report their boards have limited to no knowledge or experience with the technology⁷. NACD’s 2025 survey shows 62% of boards now dedicate agenda time to AI⁸ — but discussion is not governance, and reporting on that data indicates only about 27% have formally added AI to committee charters⁹.

Meanwhile the exposure is compounding. Gravitee’s April 2026 survey of 750 senior technology leaders found enterprise AI agent estates roughly doubled in four months while monitoring coverage, accountability structures, and pre-deployment controls barely moved — organizations becoming more comfortable with a risk they have not actually reduced¹⁰.

Here is the test I put to leadership teams: open your enterprise risk register and your governance documents. Is “an autonomous agent makes an irreversible change to a production system” written down as a loss scenario? Is a specific executive named as its owner, at a level senior enough to halt deployments? Does the document state how the risk will be mitigated and when the CEO and board get notified? If the answer is no, your governance reflects the AI you had in 2023, not the AI your staff could be running today.

SECTION THREE

What To Do About It — Actions to take today

This is a governance problem before it is a technology problem, so the fixes start with authority, in writing.

• Name one accountable executive. A CIO, CISO, CRO, or Chief AI Officer with explicit ownership of agentic AI risk, a board-visible charter, and the authority to halt any deployment. Embed AI oversight in an existing committee charter — audit or risk — rather than leaving it homeless. McKinsey’s review of the Fortune 100 found only 39% disclosed any form of board-level AI oversight⁹.
• Publish a signed authorization policy. Define autonomy tiers: actions agents may take freely, actions requiring human approval, and actions that are prohibited outright — production changes, payments above a threshold, customer-facing commitments. The accountable executive signs it; the board risk committee ratifies it; every employee who can configure an agent acknowledges it. The NIST AI Risk Management Framework’s Govern–Map–Measure–Manage structure is a sound, freely available scaffold¹¹.
• Engineer the guardrails. Separate development from production. Give each agent its own least-privilege identity — no shared credentials. Log every agent action to an immutable audit trail. Build a kill switch and test it like a fire drill. Define the rollback procedure and the notification thresholds that put an incident in front of the C-suite within hours, not weeks.
• Enforce it — and make the deterrent real. Treat flipping an agent to “always approve” outside policy the way you treat sharing credentials: a violation with consequences. Access revocation, formal discipline, and annual compliance attestation give the policy teeth. Audit periodically for shadow agents running outside the inventory. For organizations touching the EU, this is no longer optional: the EU AI Act mandates human oversight for high-risk systems and carries penalties up to €35 million or 7% of global turnover¹².

Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027 — with inadequate risk controls among the primary drivers¹³. The organizations that thrive with agents will not be the ones that moved fastest. They will be the ones that knew, before the first incident, exactly who was in charge, what the agent was allowed to touch, and who got the call when something broke.

That is not caution. That is clarity — and clarity is what lets you move with confidence.
If you need help working on these issues or would like to talk about a free assessment of  your AI Governance readiness, give me a call directly at (808) 207-2477. 

SOURCES

1. Recorded Future, “Emerging Enterprise Security Risks of AI” (citing Gartner), 2026 — https://www.recordedfuture.com/research/emerging-enterprise-security-risks-of-ai
2. Deloitte, “Agentic AI is scaling faster than governance guardrails,” 2026 State of AI in the Enterprise — https://www.deloitte.com/us/en/insights/topics/emerging-technologies/ai-agents-scaling-faster.html
3. AI Incident Database, Incident 1152: LLM-Driven Replit Agent Executed Unauthorized Destructive Commands During Code Freeze — https://incidentdatabase.ai/cite/1152/
4. eWeek, “AI Agent Wipes Production Database, Then Lies About It,” July 2025 — https://www.eweek.com/news/replit-ai-coding-assistant-failure/
5. Writer research, as reported in “AI Governance Statistics 2026,” Evolvance Market Research — https://evolvancemarketresearch.com/statistics/ai-governance-statistics/
6. Protiviti & BoardProspects, 2026 Global Board Governance Survey — https://www.prnewswire.com/news-releases/only-26-of-directors-discuss-ai-at-every-board-meeting-global-survey-finds-302714274.html
7. Deloitte Global Boardroom Program, “Governance of AI: A critical imperative for today’s boards,” 2nd edition — https://www.deloitte.com/global/en/issues/trust/progress-on-ai-in-the-boardroom-but-room-to-accelerate.html
8. NACD, 2025 Public Company Board Practices & Oversight Survey — https://www.nacdonline.org/all-governance/governance-resources/governance-surveys/surveys-benchmarking/2025-public-company-board-practices–oversight-survey/2025-board-practices-oversight-ai/
9. Forbes Business Councils, “The Boardroom’s AI Awakening” (citing NACD and McKinsey Fortune 100 analysis), 2026 — https://councils.forbes.com/blog/the-boardrooms-ai-awakening
10. Gravitee, “State of AI Agent Security Report,” April 2026 — https://www.gravitee.io/state-of-ai-agent-security
11. NIST, AI Risk Management Framework (AI 100-1), January 2023; overview via Diligent — https://www.diligent.com/resources/blog/nist-ai-risk-management-framework
12. Trustible, “AI Governance Frameworks: NIST AI RMF, EU AI Act, and ISO 42001 Compared,” 2026 — https://trustible.ai/post/ai-governance-frameworks-compared/
13. Gartner prediction, as reported in “Enterprise AI Agents Adoption Statistics 2026” — https://paul-okhrem.com/enterprise-ai-agents-statistics-2026/