Walk into any security leadership offsite this year and you’ll find the same three conversations keep surfacing for chief information security officers (CISOs). Boards have spent the year asking why the business isn’t moving faster on AI. Engineering teams are increasingly deploying AI agents faster than security processes can keep pace. And in the background, a decade of deferred patching and legacy infrastructure has spent the second half of the year groaning under the weight of it all.
These aren’t three separate problems. They’re the same problem wearing different hats, and as 2026 draws to a close, the connection is becoming harder to ignore. The lessons from this year should also shape how security leaders think about their roadmaps for 2027.
The hype has a bill, and it is coming due
For the past two years, much of the cybersecurity conversation around AI has focused on the attacker side of the ledger: deepfakes, AI-written phishing, model poisoning and increasingly automated attacks. Those threats are real. But the more uncomfortable area emerging is what rapid AI adoption is doing to organisations from the inside.
Gartner said almost 50% of cybersecurity leaders now identify technical debt as a major barrier to delivering their priorities. That is not a new complaint, but the trigger is new. Executive pressure to deploy generative AI means organisations are connecting models to unstructured data repositories and legacy systems that were never designed with AI-ready governance in mind.
The result is greater risk of data oversharing, unauthorised access and identity sprawl that has nothing to do with sophisticated attackers and everything to do with plumbing nobody wanted to fix.
At the same time, frontier AI models are increasingly being used by vendors and open-source maintainers to find vulnerabilities in their own software at machine speed. That sounds like good news: more bugs found, faster. In practice, it creates a supply-side shock. Security teams already struggling to patch on human timelines are being faced with a volume and frequency of disclosures that traditional manual processes cannot absorb. “How fast can we patch” has quietly become the wrong question. The real constraint isn’t patching capacity, it’s testing and change management.
Organisations cannot validate and deploy every fix as quickly as vulnerabilities can now be surfaced without creating the possibility of operational disruption. In the second half of this year, expect more organisations to pivot towards continuous exposure management, prioritising by what’s actually exploitable rather than working through a ticket queue by volume.
Agentic AI is already inside the building
If shadow IT was the great unmanaged risk of the 2010s, shadow AI agents are shaping up to be its 2026 successor, except this version can take actions, not just store data.
Employees are using no-code and low-code platforms to build automations, while business and application teams are connecting home-grown chatbots to internal APIs and tools. Some of those agents are operating with far more autonomy than security teams realise.
Gartner found only 16% of organisations have successfully enforced workflows governing employees’ use of third-party AI applications. That leaves plenty of room for “rogue automation”: agents running in the background with excessive privileges, sometimes long after the person who created them has stopped paying attention. This is where identity and access management (IAM) is being forced to evolve quickly.
IAM tooling designed primarily around people, static credentials and broad roles was not built for rapidly proliferating autonomous agents. Machine identity volumes have exploded over the past year, and a striking number of organisations have already had a security incident tied to a compromised AI-linked identity, with real financial and reputational fallout attached. The fix isn’t more dashboards. It’s treating every agent like a contractor: give it a name, a documented purpose, an accountable human owner, and credentials that expire on a schedule rather than living forever.
Stop trying to own everything
There is a leadership thread running underneath all of this, and it may be the more important one.
As AI and resilience expectations pile onto the CISO role, there is a temptation to say yes to everything: data governance, AI governance, privacy, digital resilience and whatever adjacent risk lands next. Because saying yes can look like influence. It isn’t. It’s a fast route to burnout and a team that owns risk it doesn’t have the resources to actually manage.
The sharper move, and the one starting to separate effective CISO’s from overstretched ones, is orchestration rather than ownership. Influence the CIO, the chief data officer, the business unit heads make them own their own technical debt and their own AI governance decisions rather than quietly absorbing accountability and risk that was never cybersecurity’s to carry. It’s a less flattering job description, but it’s a far more sustainable one.
The clock most boards still are not watching
Away from the AI noise is another, slower-moving example of the same underlying problem: post-quantum cryptography.
CISOs are being pushed to prepare for cryptographically relevant quantum computing around the end of this decade, while “harvest now, decrypt later” attacks create a reason to protect long-lived sensitive information today.
Yet progress remains limited. Among CISOs who have started preparing for post-quantum cryptography, fewer than a quarter have made meaningful progress on cryptographic inventories, and just 22% have conducted quantum risk assessments. The lesson is broader than quantum.
Security transformation happens on infrastructure timelines, not hype-cycle timelines. Discovering cryptography across an enterprise, upgrading protocols, working through vendor dependencies and replacing legacy systems takes years. Waiting until a threat becomes immediate is therefore not a strategy.
The through-line
Pull these trends apart and one theme sits underneath them.
Organisations are trying to bolt AI-scale ambition onto infrastructure, governance, identity and security processes built for a slower, more predictable technology environment. That mismatch, rather than any single new attack technique, is becoming the defining cybersecurity trend as we close out the year.
The organisations getting ahead of it aren’t the ones with the flashiest AI security tooling. They’re the ones treating technical debt, identity and governance as the actual foundation AI adoption sits on and being honest with the board about the fact that foundation needs investment before the house gets any bigger.
Gartner analysts are exploring the changing cybersecurity landscape at the Gartner Security & Risk Management Summit in London, from 22–24 September 2026
Arthur Sivanathan
Arthur Sivanathan is Senior Director Analyst at Gartner, where he specialises in the role and priorities of the CISO. His research covers security strategy, organisational structures, leadership communications and security management practices, including the use of standards and frameworks. He also provides analysis on business continuity management and organisational resilience.
