Agentic artificial intelligence (AI) is expected to become a major new attack surface for cyber criminals and software supply chains will make applications vulnerable to attack. Attacks by AI agent “swarms” are forecast to quadruple in the coming year. Deepfakes are expected to go mainstream with significant implications for security. And finally, more companies than ever are agreeing to pay ransoms to those who’ve encrypted data or who have threatened its exfiltration.

These predictions and observations were part of a recent ISC2 Knowledge Vault webinar featuring founder and CEO of CyberEdge Group, Steve Piper. (Cyber Edge Group is responsible for the annual Cyber Threat Defense Report, the next edition of which is expected out in May 2027.)

Not surprisingly, he says AI plays a roll in all five of the cyber security predictions made during the annual roundup of information for IT security specialists and executives.

Drawing on his annual research, Piper says the number of companies being victimized by successful ransomware attacks isn’t getting worse – the number currently sits around two-thirds of those surveyed that say they’ve been victimized by a ransomware demand – but more are paying ransoms to get their data back.

In 2025, his research showed that 54 per cent of payors were able to retrieve their data. This rose to 61 per cent in 2026. He also noted that 90 per cent of organizations had rising IT budgets. Those which did have a rise in their security spending reported an average 5.6 per cent increase, “which is healthy,” he says.

“Be prepared as if it is going to happen,” he says of the ransomware threat. “We’re seeing a spike in the percentage of organizations deciding to pay the ransoms. That’s not good, even if cyber insurance policies pay,” he adds. “We don’t want to fund this multi-billion-dollar ransomware industry.”

All told, Piper provided webinar attendees with 10 predictions related to cyber security which expects to come to pass in the coming year.

1. AI will begin reshaping the cybersecurity workforce but won’t reduce overall IT staffing.

“I’m not making any claim for 2028 or beyond,” he notes.

Piper says it’s not uncommon for 80 per cent of organizations to have a backlog of unfilled positions, including those for security architects, for example. He says AI may allow teams to accomplish more with fewer resources but that the need to secure AI infrastructure will in fact create new jobs. “It absolutely will,” he says. “I see a big need here.”

He notes again that AI will help improve productivity, but that attack surfaces which make a company’s systems and data vulnerable, are expanding at a rapid pace. “AI agents are coming out, all over the place,” he points out.

2. The number of successful cyber attacks by AI agent swarms will at least quadruple in 2027.

“AI agents that are coordinating together and they’re messaging each other and they all have different assignments; we call this a swarm. That’s a new vocabulary term. Please remember it,” he says. “My prediction is the number of successful cyber attacks by AI agent swarms will at least quadruple. I’m not talking about the accidental ones.”

His description of an AI agent swarm, where agents each have a role to play (“multiple agents operating in parallel for recon, credential discovery and exploitation” along with those responsible for orchestration, vulnerability discovery, exploit development and execution – harvesting, social engineering, phishing and privilege escalation), and where the agents are communicating with each other, is then tied to the concern about the damage which can be caused by a compromised agent with escalated privileges.

3. Agentic AI will become a major new attack surface.

Companies deploying AI agents – both those that are sanctioned by the company’s IT department and “shadow AI” which is the introduction of AI by employees without a security team’s vetting – may cause security teams to struggle.

Different uses for AI agents, including in customer service applications, IT help, sales, personal assistance, procurement, legal and contract review, are an endless supply of points which need to be secured. “Sometimes these AI agents are also overprivileged,” he adds. “We’re not practicing the least access to systems, we’re over-privileging these agents, which of course causes a security risk if they’re compromised.

“AI agents are targeted for prompt injection, or agent-to-agent communication, to create new attack pathways. Autonomous decision making will amplify the consequences of security breaches. If you compromise an AI agent and it’s already programmed to do things without human permission first, that can be very dangerous. Traditional security controls may struggle to govern agentic behaviour,” he says.

4. AI security will remain the fastest-growing cybersecurity market segment.

Piper notes that one out of every nine providers of cyber security solutions have at least one which addresses AI security. “Gartner forecasts $4.8 billion in AI security spending next year, up from $2.8 billion this year,” he adds. (Figures in U.S. dollars.) “That equates to a 69 per cent year-over-year increase.” Most of the categories that organizations are spending this money on, he notes, were categories that for the most part didn’t exist five years ago.

5. Deepfake fraud attacks will become mainstream.

Although deepfakes currently are something the cyber expert agrees he only sees now and again, he says by this time next year he expects deepfake attacks will be a normal thing for organizations to deal with. “This just underscores the importance of investing in your human firewalls. The big obstacle standing in the way of success is low security awareness among employees.”

While companies in the past may have done well by training their employees to detect and avoid clicking or opening malicious email links and attachments, today’s deepfakes and executive impersonation efforts are more convincing than ever. “It’s real-time deception. Sophisticated deepfakes enable attackers to impersonate trusted individuals during live video and audio calls. Attackers supplement fraudulent emails with convincing deepfake voicemails.”

Piper provides numerous examples in his webinar, including those where CEOs have been duped by deepfake discussions with imposter attorneys, attackers using cloned employee voices to trick frontline staff into resetting passwords, even staged Zoom calls with heads of state (in Singapore) which resulted in the release of millions in funds to attackers. “Someone fell for it because they were following instructions from their Prime Minister and other officials that were deepfakes,” he points out, adding that the technology has been used to sway elections in more than one country.

“There are categories of technologies to help mitigate the risk of deepfake fraud,” he says, listing deepfake detection softwares, biometric, anti-spoofing, identity verification, document authentication, real-time communication fraud prevention, payment fraud prevention and others.

“Please take note of these because this is my prediction: It’s going to get worse. It’s going to get mainstream. It’s more a question of when.”

Honourable mentions

Piper also shared with the group, five more predictions which don’t make the top of his list, but which he expects will come to pass in the coming year, nonetheless.

  • Data theft ransomware attacks without encryption will continue to rise. (The threat being that cyber criminals will publish the data in question, rather than bother to encrypt it.)
  • Compromised identities will become cyber criminals’ most valuable attack vector. “Attackers are increasingly logging in, rather than breaking in,” he says.
  • Cyber criminals and nation-state actors will increasingly target “internet of things” (IoT) devices, operational technologies and industrial control systems. Attacks could be akin to recent attacks in Europe which created millions in damages when supermarket chain refrigeration units were recently hacked and programmed to cook the food they were supposed to be storing.
  • Multiple disruptions will occur in software supply chains. “This proverbial iceberg is your application, but beneath it there are software suppliers, components that feed into your applications, including a lot of open-source components,” he notes. “When these components are vulnerable to attacks, your applications are vulnerable to attacks. Modern applications rely on hundreds of third-party libraries and packages. AI is accelerating these supply chain attacks, making it easier to discover these vulnerabilities and to exploit them,” he says.
  • Quantum computing is a looming risk. “We understand quantum computing, in terms of it becoming mainstream, we’re years away. But that is not stopping threat actors. They are doing what’s called harvest now, decrypt later,” Piper notes. “They’re exfiltrating your data knowing that in the next few years, at some point, there’s going to be capability driven by quantum computing to decrypt this storm.”