Most CISOs Are Not Confident About AI Security. Here Is What Separates the Ones Who Are.

Key points

• 41% of the 113 chief information security officers (CISOs, meaning the executives responsible for a company's cybersecurity) surveyed by IANS Research in April and May 2025 said they feel optimistic about managing AI security risks over the next 24 months.
• 38% of the same group expressed pessimism, a near-even split that surprised several outside analysts.
• IANS identified six organisational signals that separate confident CISOs from worried ones, all centred on leadership support rather than current technical controls.
• Analysts warn that feeling ready and actually being secure aren't the same thing.
• Third-party AI risk, meaning danger from AI tools built into vendor software rather than tools a company chose itself, was flagged as a blind spot.

What did the survey actually find?

Slightly more CISOs feel hopeful than fearful about AI security, but only just. IANS Research, a security-focused advisory firm, polled 113 CISOs and found the field almost evenly divided: 41% optimistic versus 38% pessimistic, with the rest somewhere in the middle.

The factors that predicted optimism weren't technical. CISOs felt better about the future when senior leadership understood AI risk, when someone in the organisation owned AI governance, when the CISO personally controlled the AI security budget, when the security team used AI tools effectively in its own work, and when staffing was sufficient. The survey found that confidence tracks with organisational power and support, not with how mature the company's security controls already are. Our story from 3 September, "The CISO Role Is Broken. A New Title Won't Fix It.", found the same structural problem from a different angle: the org chart constrains the person before the person can act.

Should that optimism be trusted?

Several analysts urge caution. Rock Lambros, director of AI standards and governance at security firm Zenity, points out a built-in survey problem: a CISO who rates their own programme as efficient is unlikely to rate their own risk as critical two questions later. People stay consistent with themselves, which can make results look rosier than the underlying reality.

Sanchit Vir Gogia of Greyhound Research draws a sharper line. Leadership support and clear governance show whether a CISO has room to act. They don't show whether the AI systems running inside the company are under control. Justin Greis, CEO of consulting firm Acceligence, put it more directly: the idea that almost half of CISOs feel optimistic about AI security doesn't match what enterprise security leaders are telling him privately. He described them as white-knuckling their way through the fastest-moving security challenge of their careers.

What are the real gaps?

Pearl Almeida of Info-Tech Research Group identifies two problems that cut across all six IANS factors.

First, most senior leaders, including security leaders, can't explain how an AI agent (a piece of software that takes actions on its own, deciding which tools to call and what to do with the results) actually works. If you can't describe the mechanism, she argues, you can't correctly size the risk.

Second, governance ownership means little if security is bolted on after the fact. Naming an owner only works when that owner accepts real accountability, up to and including security staff pausing their workflow during an incident.

Brian Levine of FormerGov adds a third gap: third-party AI risk. Many companies are running AI models they didn't choose, pulled in through vendor software or used by staff without approval. Governance frameworks that focus only on internally built AI miss this entirely.

Common questions

Does this affect ordinary employees or customers?

Yes, indirectly. When companies rush AI tools into use without proper controls, data exposures become more likely. Customers and employees whose data those companies hold carry the downstream risk.

What should organisations actually do right now?

Map every AI tool in use, including those inside vendor products, identify what data each tool can reach, and assign someone with genuine authority and accountability to oversee each one.