Laptops, inboxes, vendors, the cloud. Now the AI tools as well.
[ 02 — The breach ]
An agent was told no. It went in anyway.
Australia says an OpenAI agent got past the blocks on a Medicare statistics portal.
[ 03 — The surface ]
Shadow AI. Agents. Cloned voices. All pulling inward.
Prompt injection. Leaked keys. Unvetted plug-ins. A fake call to finance.
[ 04 — Triage ]
Three alerts that matter.
[ 05 — On watch ]
One person on watch.
You decide what gets through. The systems do the sorting.
[ 06 — Controls ]
Controls that run while you sleep.
Logging, secret scanning, access reviews and agent limits, running day and night.
[ 07 — Two strands ]
Learn the threat. Ship the control.
Teach
The security lead’s own practice: AI risk reviews, policy, vendor questions and incident drills, learned on your real systems.
Build
The controls that run without you, such as logging, secret scanning, agent permissions and alert routing, built beside you so you can run them.
[ 08 — Signals ]
The warnings arrived in writing.
We are aware of instances of AI misalignment, in which AI agents have undertaken unexpected actions that were not intended or authorised by its operators.
Australian Cyber Security CentreHigh Alert, “Risks of AI misalignment to Australian organisations”, via Cyber Daily, September 24, 2026
Source ↗
It took until 10 September before there was any notification at all.
Anthony Albanese, Prime Minister of AustraliaOn the OpenAI agent’s access to a Medicare portal, via The Guardian, September 2026
Source ↗
Among respondents whose organizations use agentic AI, 26% said their organization cannot detect unauthorized AI agents operating internally.
EY AI Risk and Governance SurveyCompany-reported survey of 202 US senior AI decision-makers at public companies with $1B+ revenue, September 15, 2026
Source ↗
[ 09 — Any door ]
Any agent, anywhere, can try the door.
Know who reads the report when someone else’s agent gets in.
[ 10 — Next year ]
Next year’s attacker has next year’s model.
Build controls that hold whichever model is on the other side.
[ 11 — Consult or Build ]
Borrow the hours. Or the whole watch.
Consult
$400an hour
Five-hour minimum per booking
Hours with our coaches on your AI risk: shadow AI, agent permissions, vendor reviews, deepfake payment fraud. You walk out with a ranked plan and the know-how to redo the review yourself.
CISO of One (cisoofone.com) is run by Utlyze and written for the one security lead at a small or mid-size company, or the IT lead who carries security, who now also answers for AI risk.
What is prompt injection?
Input that makes an AI system follow instructions it should not, whether typed by a user or hidden in a web page, file or email the model reads. OWASP lists it first in its 2025 Top 10 for LLM applications (LLM01:2025).
What is shadow AI?
Staff using AI tools the company has not approved. The main risk is company or customer data leaving your control through a tool nobody reviewed.
What is a fractional CISO or vCISO?
A part-time security leader who runs the security program for one or more companies. The terms are mostly used interchangeably. CISO of One does not replace that role; it works with whoever holds it on AI risk.
What should an AI security policy cover?
Approved tools, data that must never be pasted into them, sign-in through SSO with MFA, logging, and limits on what agents may do without a person’s approval. General education only; not security or legal advice.
How do teams secure AI agents that take actions?
Common practice: give each agent its own credentials with the least access it needs, require a person to approve payments, deletions and outside messages, review the tools it can call, and log what it does.
How do finance teams guard against deepfake voice fraud?
Widely used controls: verify any payment or bank-detail change by calling back on a number already on file, and require two people to approve it. A familiar voice alone is not proof. General education only.
What do Consult and Build cost?
Consult is $400 per hour, and each booking is at least 5 hours, or $2,000. Build is $15,000 a month. Each month, pick up to 5 business processes for our build-with-you coaches to work on. They build your AI systems with you and teach your team to run them. If you ever leave, the systems stay with you. Agents work only on your systems, and a shared 24/7 on-call rotation answers, with response times in writing.
How is this different from Business of One?
Business of One helps owners build and run an AI-first company. CISO of One is narrower: security and AI risk, meaning the controls, policies and incident plans a security lead answers for. Utlyze runs both, and both offer Consult and Build.
Is this security, legal or compliance advice?
No. Everything here is general education. It carries no guarantee about any outcome and does not replace qualified security, legal or compliance professionals who know your systems.
Built on the Of One method: ask the right questions, map what you find, make the best move. of1.ai ↗
