Advancing the Safety
of AI Agents.
We are an independent UK research organisation investigating the safety, security, and verifiability of AI agents and tools. Our research is closed but may be made available for licence or as open source in the future.
What We Research
Our research programme covers five interconnected areas of AI safety and security which all currently closed.
Model Package Protocol (MPP)
A licensed protocol for secure, signed, and sandboxed AI tool artifacts, commercialised by Quantum 2x. MPP brings container-like isolation to AI agent tool execution with WASM sandboxing, Ed25519 signing, and fine-grained permission boundaries.
Behaviour Verification for AI Models
Research into methods for formally and empirically verifying the behavioural properties of AI models, ensuring they act within defined, auditable bounds and do not exhibit unintended emergent behaviours.
AI Agent Review & Monitoring
Systems for the continuous review and real-time monitoring of AI agents in operation, tracking decisions, tool invocations, and data access with full traceability and anomaly detection.
Security Protocols for Agents in Production
Research into security architecture and protocol design for AI agents deployed in live environments, covering isolation, least-privilege execution, inter-agent trust, and supply-chain integrity.
AI Agent Audit Tools
Tools and frameworks for auditing AI agent behaviour post-hoc, enabling compliance teams, researchers, and operators to inspect, replay, and assess agent decisions with cryptographic integrity guarantees.
Research-First Thinking for the AI Era
AI agents are becoming autonomous decision-makers with access to sensitive systems. Understanding and verifying their behaviour before and during deployment is a foundational challenge for the field.
Quantum 2x conducts independent research into the safety, integrity, and auditability of AI agent systems from formal behaviour verification to production security protocols.
Get in Touch