CRAIG ALLAN-MCWILLIAMS

I build security for AI systems that make decisions and take actions without someone watching over their shoulder. The industry calls this agentic AI. I think it is one of the least-understood and least-protected attack surfaces in the modern enterprise, and I have been working on it long enough to have opinions.

My flagship open-source project is raucle, released under Apache-2.0. Raucle produces a cryptographically verifiable capability receipt for every action an AI agent takes through the platform. Each receipt is signed, content-addressed, and independently verifiable by any third party without contacting a vendor. The goal is practical: help regulated organisations prove what an agent did, under whose authority, against which policy, and with what result. The repository is at github.com/craigamcw/raucle, and the work is under submission at IEEE Security & Privacy 2027.

Raucle covers agent identity, delegated authority, portable trust, revocation, cross-organisation handshakes, provenance chains, and compliance evidence packs. It is not a detection tool with audit bolted on. The audit is the product.

The questions I work on are not theoretical:

• How should trust boundaries work for autonomous agents?
• How do you audit agent decisions in real time?
• How do you stop an AI system becoming a lateral-movement vector into critical infrastructure?
• How does an organisation prove that an agent acted within its authority and policy?
• What security controls do you need when agents can call tools, access data, and trigger workflows on their own?

Organisations are already deploying agents into systems that hold sensitive data, privileged identities, and operational authority. These questions are not future concerns. They are current ones.

I have spent over 25 years as a security engineer, architect, consultant, and adviser across banking, government, global enterprise, cloud platforms, and technology delivery. That includes CI/CD pipeline security, AWS security architecture at scale, security monitoring and forensics, enterprise risk management, application and cloud security, and security programmes built from nothing up to operational.

I hold SABSA Chartered Security Architect, CISSP, AWS Certified Security Specialty, CCSK, and MITRE ATT&CK credentials. My approach treats security as a system property: something designed into architecture, identity, software delivery, operations, governance, and organisational decision-making. Not a control layer added at the end.

I bring both sides of the problem. The technical depth to architect and engineer secure infrastructure, and the commercial fluency to explain risk, trade-offs, and investment decisions to boards and executive teams. Security architecture that an executive cannot understand is architecture that does not get funded.

I write about cybersecurity and AI security at All Things on Security, covering agentic AI, AI governance as architecture, threat modelling, fail-open vulnerabilities, quantum security, and securing AI across the software-delivery lifecycle.

Areas of focus

• Agentic AI and AI-agent security
• Agent identity, authority, and authorisation
• Cryptographic provenance and accountability
• Security architecture and SABSA
• AWS and cloud security
• CI/CD and DevSecOps security
• Threat modelling and adversarial testing
• AI governance, assurance, and audit evidence
• Enterprise cyber risk and security transformation
• Open-source security engineering