Fernando Montenegro serves as the Vice President & Practice Lead for Cybersecurity & Resilience at The Futurum Group. In this role, he leads the development and execution of the Cybersecurity research agenda, working closely with the team to drive the practice’s growth. His research focuses on addressing critical topics in modern cybersecurity. These include the multifaceted role of AI in cybersecurity, strategies for managing an ever-expanding attack surface, and the evolution of cybersecurity architectures toward more platform-oriented solutions.
Before joining The Futurum Group, Fernando held senior industry analyst roles at Omdia, S&P Global, and 451 Research. His career also includes diverse roles in customer support, security, IT operations, professional services, and sales engineering. He has worked with pioneering Internet Service Providers, established security vendors, and startups across North and South America.
Fernando holds a Bachelor’s degree in Computer Science from Universidade Federal do Rio Grande do Sul in Brazil and various industry certifications. Although he is originally from Brazil, he has been based in Toronto, Canada, for many years.
Predictions
Agentic AI Identity Sprawl will Break Traditional IAM, Requiring Dedicated Authorization Governance to Prevent Unmonitored Agent Activity
As 2026 ends, enterprises will realize their Identity and Access Management systems are failing because they were built for humans, not goal-directed AI agents. Standard workload identity models designed for service accounts simply do not work for these new entities.
Meanwhile, agent identity sprawl will continue to accelerate due to unmonitored procurement, unmanaged open-source deployments, and agents embedded within SaaS platforms. Organizations will need to urgently build agent authorization governance fit for purpose. Without it, the costly gap between an agent’s authorized capabilities and its actual actions will become a severe liability.
Why This Is Trending
Agent Taxonomy Is Exposing Governance Blind Spots: Not all agents are the same problem. End-user-facing agents, such as copilots and assistants operating on behalf of individuals, carry different risk profiles than application-layer agents orchestrating workflows across systems, which differ again from agents embedded inside third-party SaaS platforms that security teams never directly reviewed. Each category has distinct identity models, oversight frequencies, and blast radii. Organizations that govern agents as a single, uniform category systematically under-engineer controls for autonomous, goal-directed agents while over-engineering them for bounded, deterministic ones.
Credentials Solve Authentication, Not Authorization: Workload identity standards give agents a verifiable identity. They do not solve the runtime authorization gap, specifically, whether a specific action combining steps across multiple systems falls within what was actually intended when a human granted a goal. Goal-directed agents infer permission from objectives; they do not seek explicit authorization for discrete actions. When an agent causes harm, “Who authorized this?” has no clean answer. Logging and PAM provide detective value but cannot substitute for a prospective authorization layer that does not yet exist at enterprise scale.
Third-Party Agents Are an Information Asymmetry Problem: Agent sprawl is arriving faster than shadow IT ever did, partly because it travels through legitimate procurement channels. A vendor platform approved by a marketing team may contain embedded agents with broad data access that nobody in security has evaluated. There is no established SBOM equivalent for agents, meaning there is no standard disclosure mechanism specifying which agents a product contains, what permissions they require, and which systems they touch. Buyers cannot assess what they are deploying, and vendors have no market incentive to disclose more than necessary.
Vendor Impact
Agent Taxonomy as a Governance Starting Point: Security teams are mapping agent environments by category, covering end-user agents, application orchestrators, and embedded third-party agents, and applying governance proportional to autonomy and blast radius. This bottom-up approach, starting with the specific runtimes already running in the organization, surfaces the embedded third-party agent population that would otherwise be invisible and yields more actionable results than waiting for a top-down framework.
Extending PAM and Requiring Agent Capability Disclosure: For high-value, known target systems, extending PAM to cover agent identities is the right near-term control, with honest acknowledgment of where it breaks for agents that discover APIs at runtime. For third-party agents, organizations are beginning to require capability disclosure in vendor contracts as a procurement forcing function, establishing what an agent can access before deployment rather than after an incident.
Token Spend as a Trust and Governance Signal: Runaway agent spend is becoming a governance concern in its own right. An agent with broad authorization and unconstrained tool access can consume compute and API budget well beyond what was anticipated at deployment. Organizations experiencing unexpected cost spikes are often those with the weakest authorization controls. Security and finance teams are beginning to treat the token budget as an authorization dimension: an agent that cannot account for its spend cannot be trusted with broad permissions.