Security Experts Urge Redesign of Trust Architectures for AI Agents

Security experts are calling for a fundamental redesign of trust architectures to mitigate risks from agentic AI and quantum computing. Pravir Malik, founder of QIQuantum, argues that traditional security is insufficient because AI agents can execute actions at machine speed and potentially exceed their intended authority. He proposes a new trust category based on quantum no-go laws, utilizing a non-exportable physical quantum object as the root of trust.

John Sahlin of General Dynamics Information Technology warns that AI agents function as digital insiders capable of bypassing traditional guardrails. He advocates for an evolved Zero Trust model where agents possess no persistent privileges and instead receive temporary, conditional access for specific tasks. Both experts emphasize that organizations must secure underlying architecture and data provenance before deploying autonomous agents to prevent the chaining of low-criticality vulnerabilities into high-impact breaches.

These warnings follow broader industry shifts toward post-quantum security. The National Institute of Standards and Technology finalized post-quantum cryptography standards in 2024 and launched the 2026 AI Agent Standards Initiative. In response to these threats, Cloudflare has implemented post-quantum encryption for over half of its human-initiated traffic, while Microsoft developed the Pluton security processor and Microsoft Entra Agent ID to provide AI agents with purpose-built identities.