AI Agents Have an Identity Complex With Jeff Malnick

2026-05-20 · Source: AI Explained · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Software Development & Engineering · Depth: Advanced, extended

Summary

Jeff Malnick, VP of Engineering at 1Password, highlights the urgent need for robust AI agent identity solutions, as traditional OAuth flows and static permissions fail to address agents' dynamic reasoning capabilities. He explains that current systems often lead to 90-95% over-privileging, a risk amplified by agents' ability to request new permissions dynamically. 1Password proposes a framework of three agent identity models—delegated, bounded, and fully autonomous—each with local and remote variants. Securing local environments like laptops is particularly challenging, as file system access can expose clear-text secrets and recovery codes, enabling agents to bypass security. Malnick advocates against using bearer tokens and passwords for agents, instead recommending short-lived access tokens and dynamic, just-in-time authorization that pulls human intent through policy decision and enforcement points. He also emphasizes that credentials must never enter LLM context.

Key takeaway

For AI Security Engineers designing agentic systems, you must implement dynamic, just-in-time access controls, ensuring credentials never enter LLM context and utilizing short-lived access tokens. Prioritize building applications with granular scoping from the outset and consider proxy-based token injection to mitigate risks like prompt injection and over-privileging, especially for agents operating in less controlled local environments.

Key insights

AI agents' reasoning capabilities necessitate dynamic, just-in-time identity and authorization beyond traditional static permission models.

Principles

• Delegate permissions dynamically.
• Never put credentials in LLM context.
• Prioritize short-lived access tokens.

Method

1Password uses OS-level attestation via kernel audits and code signing to establish a chain of trust for local agent processes, enabling secure identification and policy application.

In practice

• Implement proxies for token injection.
• Build applications with domain-level scoping.
• Utilize Workload Identity Federation.

Topics

• AI Agent Security
• Agent Identity Management
• Dynamic Authorization
• Secrets Management
• Prompt Injection
• Workload Identity Federation

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Security Engineer, AI Engineer, MLOps Engineer

Related on AIssential

• Identiverse 2026 Recap: Identity Security For Agentic AI Dominates — Identity security must rapidly adapt to govern autonomous AI agents and nonhuman identities with real-time, context-aware decisions.
• AI Privilege Escalation: Agentic Identity & Prompt Injection Risks — AI privilege escalation exploits agent over-permission, inheritance, prompt injection, and misconfiguration, necessitating robust mitigation strategies.
• Agentic Trust: Securing AI Interactions with Tokens & Delegation — Securing agentic AI systems requires robust identity, authentication, authorization, and secure propagation mechanisms.
• Identity for AI Agents - Patrick Riley & Carlos Galan, Auth0 — New Auth0 features enable secure identity and fine-grained authorization for AI agents, addressing emerging security challenges.
• The Rise of the Machine Identity: Securing the AI Workforce and AI Agents — AI agents introduce exponential security risks due to their scale, autonomy, and ephemeral nature, demanding new defense strategies.
• The Identity Crisis of AI Agents: Why Autonomous Systems Need IAM Before They Need More… — AI agents require unique, verifiable identities and continuous trust evaluation to prevent an enterprise security crisis.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Explained.