Industry executives and experts share their predictions for 2026. Read them in this 18th annual VMblog.com series exclusive.
By Ravi Ithal, Chief Product and Technology Officer, AI Security, Proofpoint
AI seems like a part of every 2026 prediction, but it’s for good reason, especially in the cybersecurity space. From cyberattacks that leverage an AI website generation platform to agentic AI solutions that improve how organizations detect and mitigate compliance risks, AI is increasingly shaping both the opportunities and risks for this industry.
AI agents have become teammates with whom security professionals can automate collaboration and data security. Given this new technology and dynamic, we need to rethink protections and how to shift from human-centric to human and AI-centric security. As we allow agents to take autonomous action, proper levels of control must be in place to ensure they’re not introducing more risks to the organization, while at the same time providing unparalleled efficiency and additional security.
Let’s explore what 2026 might have in store, for AI and beyond.
AI Agents Will Become the New Insider Threat
By 2026, autonomous copilots may surpass humans as the primary source of data leaks. Enterprises are rushing to roll out AI assistants without realizing they inherit the same data hygiene issues already present in their environments. Over-permissioned SharePoint folders, unclassified documents, and outdated access rules will allow these copilots to surface sensitive data to users who were never meant to see it.
These agents are not simply tools; they will become identities in their own right, with each one carrying a trust score, behaving as a peer actor in the ecosystem. The old model of phishing will be replaced by “prompt paths,” or avenues through which an agent is tricked or mis-led into extracting and exposing data. Security teams will no longer focus solely on human actors; they will be forced to treat their AI agents as first-class identities, managing their privileges, monitoring their behaviors, and scoring their risks.
Model Poisoning and Prompt Injection Will Go Mainstream
We are entering a phase where the true front line of attack will no longer be firewalls or endpoints – it will be the training pipelines. Adversaries will weaponize corrupted datasets as the new backdoor, subtly poisoned data feeding into models so that downstream behavior is compromised. As a result, classification will move from simply tagging files to classifying ML supply-chains for integrity and trustworthiness. Security leaders must begin viewing the training pipeline as a critical control point and apply supply-chain standards to the data that feeds every model, every prompt, and every agent.
Shadow MCPs Will Proliferate Across Enterprises?
In the near term, we will witness the rise of unmanaged local Model Context Protocol (MCP) servers acting like the next version of Shadow IT. Rogue agents or model endpoints will spin up, connect into public LLMs, and start ingesting live enterprise data outside any governance or visibility. At that point, detection alone will become insufficient; organizations will need to auto-register every agent through a Secure Agent Gateway before it gains access. The concept of “bring your own model” will morph into “bring your ungoverned model” unless governance mechanisms evolve to force auto-enrollment and real-time trust scoring of every MCP that touches corporate data.
Looking Ahead to 2026
The cybersecurity landscape is evolving rapidly. In 2026, the rise of AI agents as insider threats, the heightened vulnerability of training pipelines, and the prominence of shadow MCPs will require a new level of security preparedness.
To thrive in this new era, those in the cybersecurity space must create a culture that is both human and agent-centric. AI agents will act as collaborators that amplify human capabilities, but they will also introduce new risks that require steady governance. Security professionals must build frameworks that foster collaboration and diligence, ensuring that individuals and AI agents work in harmony to safeguard the agentic workspace.
By embracing innovation and staying nimble and resilient, we can rise to meet the moment.
##
ABOUT THE AUTHOR
With a career steeped in enterprise and cloud security, Ravi Ithal is the Chief Product and Technology Officer of AI Security at Proofpoint. In this role, he leads the company’s Agentic Security products roadmap and strategy.
Before joining Proofpoint through the acquisition of Normalyze, a leading DSPM company he co-founded, Mr. Ithal built a reputation as a trailblazer in the security industry. As the co-founder and chief architect of Netskope, he helped develop cloud-native solutions to defend businesses against modern threats while protecting critical data. Mr. Ithal was one of the founding engineers at Palo Alto Networks, shaping the architecture behind one of its security platforms. His earlier career saw him holding impactful engineering roles at Juniper Networks and Cisco, where he honed his skills in building secure, scalable systems.
