AI Is Expanding Cyber Threats and Tightening Cybersecurity

Although threat actors are using AI to learn from repeated failures, organizations maintain a slight edge over attackers, Horwitz asserted. 

“We have the ability to train our models on all the different combinations of the type of mutations coming in, and that will ultimately provide us with a better set of detection capabilities,” he said.

Aimee Simpson agrees. Because they have access to the same technology stacks as criminals, she observed, enterprise cybersecurity teams are well-positioned to fight back and ultimately win.

“AI-based threats are a threat and need to be taken seriously, but we’re not battling these threats in the industry with outdated technology,” said Simpson, director of product marketing at Huntress, a cybersecurity company founded by veterans of the National Security Agency. 

“Cybersecurity defense teams also have access to AI tools, which are phenomenal at detecting these evolving threats in real time.”

In fact, leading cybersecurity organizations have already begun rolling out AI-first security systems that use AI technology within existing defense mechanisms.

“Embracing AI may be a useful approach in many organizations, but it’s only an effective choice if security teams can ensure that the products brought into a company are safe, secure and not going to present a data breach risk,” Simpson said.

Internal Enterprise AI Security Risks

If external criminals using AI are one type of risk, then internal employees using AI are another type.

Indeed, the rapid implementation of AI by enterprises also creates security risks by expanding attack surfaces, exposing sensitive data and introducing model‑specific vulnerabilities that existing controls don’t cover.

According to survey data from IT consultancy Gartner, 62% of organizations have experienced a deepfake attack involving social engineering or exploiting automated processes. Another 32% of organizations reported experiencing an attack on AI applications that leveraged the application prompt.

“As adoption accelerates, attacks leveraging [generative AI] for phishing, deepfakes and social engineering have become mainstream, while other threats — such as attacks on [generative AI] application infrastructure and prompt-based manipulations — are emerging and gaining traction,” Akif Khan, an industry analyst on Gartner's cybersecurity and AI team, said in September 2025 at the Gartner Security & Risk Management Summit in London.

Currently, only one in 10 organizations is prepared to protect itself from AI-augmented cyber threats, according to consulting giant Accenture. In a June 2025 report, it said 63% of companies worldwide lack both a cohesive cybersecurity strategy and necessary technical capabilities to defend against AI-driven threats. 

Those companies need “a fit-for-purpose security governance framework and operating model … to establish clear accountability and align AI security with regulatory and business objectives,” Accenture noted.

The importance of AI governance is particularly pronounced in the growth of shadow AI, which is the unsanctioned use of AI tools and applications by employees who are looking for solutions that save time or increase productivity, suggested Simpson. 

“Enterprises need to outline the risks of shadow AI in their organization and create clear compliance structures around AI usage,” she said.

Oleg Naumenko agreed. Because many companies aren’t managing AI access properly, employees often use AI tools without sufficient oversight, he observed. 

“Sometimes they grant access to the company's internal data through APIs without the company even knowing,” said Naumenko, CEO at Hideez, a provider of password-free authentication and identity access management (IAM) solutions.

Naumenko recommends integrating AI access management directly into IAM systems with a zero-trust security mindset. 

“Instead of letting individual employees manage AI on their own, IT admins should be the ones setting the rules,” he continued, adding that organizations should give AI systems access only to the enterprise data and infrastructure that they need, and only for a limited time.

Shingala echoes the importance of zero-trust principles, including strict access controls, visibility into what data models consume, continuous monitoring of inputs and outputs, and complete auditability. 

“AI should not be treated as an isolated project or a nice-to-have enhancement. It needs to be treated as a fundamental layer of the enterprise architecture,” he said. “Because AI touches so many systems and so much data, its security model must be equally comprehensive.”

Rather than viewing it as deploying new software, organizations should consider AI — and agentic AI, in particular — as a new class of employees, suggested Horwitz. 

“If you think about agentic AI and how it interacts, you have similar characteristics to employees,” he said. “Employees have intentions, they have autonomy and they need access to resources.”

Employees and AI also have something else in common: They can make consequential mistakes. Therefore, thinking of AI as a new kind of employee “will help organizations make better decisions,” Horwitz said.

Fighting AI with AI

In the face of external and internal risks alike, effective organizations are the ones that treat AI not as a productivity tool, but rather as a core security capability, Shingala observed. 

“Adopting AI without building an AI-aware security posture does not just improve business operations. It can unintentionally strengthen the capabilities of attackers, as well,” he said, summarizing the future of enterprise security as a contest of AI against AI.

“Attackers are using AI to scale and evolve, and defenders must rely on AI to predict, identify and neutralize threats just as quickly.”