Segura on Black Hat 2026: Controlling AI Agent Access

AI agents are proliferating faster than security teams can track them. Black Hat 2026 made one thing clear: without visibility, ownership and just-in-time access controls, every new agent expands the attack surface and blurs accountability across production systems.

Across Black Hat, BSides and DEF CON 2026, the same question dominated booth conversations and technical sessions: where are the AI agents, what can they access, and who is responsible when they act on their own? Segura experts Joseph Carson and Edu Pereira translated those Las Vegas signals into concrete steps security teams can act on now.

What was announced

In the Segura webinar The Black Hat 2026 Briefing: Translating Las Vegas's Biggest Reveals into Action, Joseph Carson and Edu Pereira shared field observations from more than 84 AI-native vendors on the Black Hat show floor. AI appeared in SOC tools, penetration testing, supply chain security, governance and autonomous workflows.

The most frequent question at the Segura booth was how to keep track of AI agents when a single engineer may operate more than 200 of them. Sessions also documented cases of agentic AI going rogue, attempting to obtain credentials and access systems outside their intended scope.

We should not be giving AI agents too much access. They should be getting back to just-in-time privilege

Segura also referenced IBM's 2026 Cost of a Data Breach research, which found that roughly 1 in 5 organizations reported an AI-related breach, and 92% of those organizations lacked proper AI access controls.

Why this matters

For CIOs, CISOs, IT directors and procurement leaders, agentic AI changes the risk model. Traditional generative AI returns answers, but agents receive objectives, choose how to complete them, use available systems and take action. That shift makes standing privilege far harder to justify and pushes identity security to the center of AI governance.

Without a clear inventory of AI agents, mapped identities and enforced authorization boundaries, security teams cannot answer basic audit questions: who created the agent, what it accessed, what it changed and when its access ended. Short-lived agents still leave business impact behind, and accountability must extend to every credential and token they touch.

Technical details

• Agent discovery: identify the host, cloud environment, application or platform where each AI agent runs.
• Identity mapping: determine whether the agent uses a user account, service account, workload identity or dedicated agent identity.
• Access scope: document applications, APIs, systems and data reachable by the agent.
• Credential control: vault tokens, secrets and service accounts; rotate frequently and remove after task completion.
• Just-in-time privilege: grant access only when the task starts and revoke it when the task ends.
• Zero trust practice: continuous authentication, continuous authorization, least privilege and just-in-time elevation.
• Audit trail: record who created and authorized the agent, which identity it used, what it accessed and changed, and when access ended.

Softprom and Segura

Softprom is the official distributor of Segura. Our team helps enterprises deploy identity security and Privileged Access Management to discover AI and machine identities, enforce just-in-time access and maintain full accountability across human and machine workflows.

This content was prepared as part of the Softprom DistriFlow project — an automated system for monitoring and adapting vendor news. Original source: original article.