Boards hiring a CISO this year need to ask a question most interview guides skip: how would you protect the AI systems this company relies on for decisions?
The reason is speed. Carlos Alberto Silva of 33N Ventures told me on the AI Builders Podcast that finding a vulnerability and building an exploit can now take hours, even for someone without deep expertise. He expects the cost to keep falling toward zero.
Why Cheaper Exploits Mean More Demand for Security Leaders
Anthropic announced Claude Mythos Preview on April 7 and kept it out of public hands after the model found thousands of zero-day vulnerabilities, including some in every major operating system and web browser. Investors had already reacted. When details leaked in March, cybersecurity stocks fell on the bet that frontier models would soon do what security vendors charge for.
I think the market read it backwards. Cheaper exploits lead to more attacks, and companies facing more attacks spend more on defense. A few security categories will shrink as AI takes over their work. I’d expect the rest of the market to keep growing.
Apono, a 33N portfolio company, shows where that spending goes. It gave human developers just-in-time access to sensitive systems and later extended those controls to AI agents. 1Password acquired the company in June 2026. Access opens when a task starts and closes when it ends. That’s the control you want once agents begin logging into production, and joint guidance from US and allied cyber agencies warns about privilege compromise and scope creep, where agents end up with more access than their tasks require.
Cheaper, faster attacks should change what boards look for in their next CISO. A security leader now faces attackers who need less time and money, and the interview has to test for it.
Why AI Decision Systems Are Now Part of the Attack Surface
Once a company lets models make decisions, the models themselves become targets. Protecting the data underneath is no longer the whole job. The attack surface grows each time another decision moves from a person to a system, and it becomes more critical to defend.
Exein shows how the threat extends beyond the data center to connected devices. The company embeds security at the firmware level and has expanded into protecting AI-enabled physical systems, from factory robots to drones. On a production line, a false positive that stops the line is its own kind of incident, so the security has to be precise. Even a compromised smart light bulb can open a path into critical infrastructure.
A CISO candidate whose answer stops at laptops and cloud accounts is defending an old perimeter.
Should the CISO Cover Physical Threats Too?
For companies that run critical infrastructure, digital and physical security are getting harder to separate. Carlos put it bluntly on the podcast:
“Every single physical attack starts on the Internet.”
I’d state it more narrowly. Attackers preparing a physical or hybrid strike often hire people or register domains first, and those steps leave digital signals someone can watch.
Osavul, another 33N-backed company, was started by Ukrainian founders after the invasion to read those signals and predict physical and hybrid attacks on critical infrastructure.
If you operate ports or power grids, the CISO brief should say whether physical and hybrid threat intelligence belongs to the security leader. Without a named owner, early warning signals can fall between the cybersecurity and physical security teams.
Who Should Own AI Agent Security?
Boards should settle this before the search starts. Agent security could sit with the CISO or with an AI leader such as a VP of AI or Head of AI. If the brief leaves it open, every candidate will answer differently, and you won’t be able to compare them.
My recommendation is to make the CISO accountable for security policy and incident response across all agents, with the CIO or AI leader running the controls inside the systems they own. Reporting lines vary from company to company. Every agent still needs a named owner and a clear escalation path.
What Should Boards Ask CISO and VP of Security Candidates About AI?
These are the four questions I’d put to any CISO or VP of Security candidate in a cybersecurity executive search right now.
1. Which AI Systems Here Make or Shape Decisions, and How Would You Find Out?
I’d expect candidates to start with an inventory of the models built in-house and the AI features inside software you already pay for, then trace where each one gets its data. Anyone who files AI under the IT asset list hasn’t thought it through.
2. How Would You Give an AI Agent Access to Production Systems?
Listen for access that expires and can be traced back to a person. Good candidates will explain how they’d revoke and log it. Standing permissions are the setup just-in-time access was built to replace.
3. How Would You Detect an AI Agent Acting Outside Its Task?
The answer should cover monitoring that compares what an agent does with its assigned task, and how fast the candidate could cut off access when the two don’t match. Candidates who know the agentic AI guidance will bring up logs that record each agent’s tool use.
4. If Exploits Can Be Built in Hours, What Changes in Your First Year?
You want specifics, like shorter patch windows or where the budget moves. “We’ll buy an AI security product” isn’t a plan.
Where This Argument Falls Short
Not every company trusts models with decisions yet. If yours doesn’t, the AI questions above matter less right now than hiring someone who can handle cheaper, faster attacks on the systems you already run.
Some parts of security are also being displaced by AI. If most of your spend sits there, the brief may call for a leader who consolidates tools before building a new AI security program.
Frequently Asked Questions About Hiring a CISO for AI Security
Did Claude Mythos Replace Cybersecurity Tools?
No. Mythos Preview showed that AI can find and exploit vulnerabilities, and cybersecurity stocks fell on the news. The selloff looks like an overreaction, since cheaper exploits create more work for defenders and only part of the market faces disruption.
Should the CISO Own AI Agent Security?
The CISO should be accountable for policy and incident response across all agents, while the CIO or AI leader runs the controls in their own systems. The board should decide this when it writes the CISO brief.
What Is Just-in-Time Access for AI Agents?
An agent gets access to a system only when a task needs it and loses that access when the work is done. Apono, which 1Password acquired in June 2026, built this for human developers before extending it to agents.
What Should a Board Ask a CISO Candidate in 2026?
Ask how they’d find the AI systems that make decisions and how they’d control what agents can access. Then ask what changes in their first year now that exploits can be built in hours.