By Dane Fiori, Founder of Guardare. Guardare were winners of the ‘AI Startup of the Year’ award at the 2025 AI Awards.
Artificial intelligence has moved quickly from promise to pressure in cybersecurity. In just a few years, AI has gone from an emerging capability to a board-level expectation. Many organizations now assume that if AI is present in their security stack, better outcomes should follow.
But as adoption accelerates, a quieter shift is happening among leaders: the conversation is moving from capability to concern. It’s no longer, “Can AI do more?” but “Can we trust what it tells us enough to act?”
That question is showing up in both the data and conference chatter.
In PwC’s 2024 Global Digital Trust Insights, 52% of respondents said they expect GenAI to lead to catastrophic cyber attacks in the next 12 months. That’s not a fringe concern; it’s a mainstream expectation shaping risk posture and governance decisions.
At the same time, leaders are also questioning whether organizations are ready for the security implications of AI adoption. Accenture research reported by Axios found that 36% of security and technology executives said AI is advancing faster than their security capabilities, and Accenture estimated that about 90% of those companies lack the security standards needed to defend against present-day AI-driven threats. This tension is the heart of the current moment: leaders are investing in AI while simultaneously worrying that the same wave will increase risk, reduce control, or create decisions they can’t defend.
The Gap Isn’t Just Adoption, But Also Trust
If you only look at adoption, the momentum is obvious. McKinsey’s 2025 global survey found that 88% of respondents said their organizations are regularly using AI in at least one business function, up from 78% the year before, though many are still in early stages of scaling. About one-third say they have begun to scale AI programs across their organizations, indicating growing but uneven progress in enterprise deployment.
But operationally, many teams are discovering that adding AI doesn’t automatically reduce uncertainty. In security, speed is valuable only if it leads to decisions that teams can stand behind.
That’s why the industry is starting to separate two very different outcomes:
- AI that accelerates action
- AI that accelerates noise
The difference is not the model. It’s the operating philosophy around how AI should influence decision-making.
Why “More Automation” Isn’t Always the Breakthrough
Cybersecurity has always been a high-stakes decision environment. Even when tools automate pieces of a response, responsibility still remains human. That means explaining risk to executives, proving due diligence, meeting regulatory expectations, and choosing what to fix first when everything looks urgent.
That’s why “automation first” approaches can backfire when they create black-box outputs.
Recent research from IBM underscores this gap. In its latest breach analysis, fewer than half of organizations (49%) said they would increase security investment
following a breach, a significant year-over-year decline. At the same time, even among organizations that report extensive use of AI and automation, only about
one-third apply these technologies across the full security lifecycle, from prevention through response.
The signal is clear: automation alone is no longer enough to inspire confidence or action. Leaders are becoming more selective, prioritizing initiatives that measurably improve how teams move from detection to decision to containment.
In practice, many organizations are discovering that purely automated “answers” create a new bottleneck, which is the human work of interpreting results, explaining impact, and standing behind decisions when it matters most.
The New Standard: AI as Decision Support, Not Decision Replacement
As a result of this shift in AI perspective, many leaders are changing what they demand from AI.
They are less impressed by “autonomous” promises and more focused on questions like:
- Can we trace how the conclusion was reached?
- Can analysts validate it with evidence, not intuition?
- Can we explain it to a board, regulator, or customer without hand-waving?
- Does this reduce time to action, or does it add a new layer of interpretation?
This is exactly where the market is starting to mature. And remember, healthy skepticism isn’t anti-AI; it’s a recognition that in cybersecurity, confidence can be an outcome, but not a feature.
That’s why a different AI philosophy is emerging: AI should function as a visibility engine, not a mystery machine.
What Good AI Looks Like in Cybersecurity
Across the industry, the strongest AI narratives are shifting toward a few consistent principles:
- Transparency over opacity – AI should surface the “why,” not just the “what.”
- Correlation over isolated scoring – Risk rarely lives in one place; usually it grows through relationships.
- Human control over forced automation – AI should propose, prioritize, and explain, but leave humans responsible for final decisions.
- Traceability over trust-me outputs – If teams can’t validate the path from signal to conclusion, they won’t act on it.
These principles aren’t limited to security tools or teams themselves. They’re also coming up more often in broader organizational discussions. When leaders talk
about potential GenAI-driven incidents, it actually reflects a broader concern about how quickly AI is scaling relative to existing controls. And when executives note that AI adoption is moving faster than security readiness, it reinforces a growing awareness that progress depends on pairing innovation with oversight and clear operational standards.
About the Author: Dane Fiori
