Research & publications
Independent work by Junior Williams, with published papers, supporting evidence, and plans for further study.
- Author: Junior Williams
Recent insights and operational guidance
- Post-Quantum Cryptography in Canada: Requirements, Standards and Vendor Readiness
- Before You Buy Another AI Security Tool, Map What You Already Have
- Changing the Model Means Retesting the Workflow
- Who Gave the Agent Permission?
- Check Permission at the Moment of Action
- CPCSC Level 1: Build an Evidence File You Can Explain
- What to Ask an AI Vendor After the Demo
- The AIA Starts Before the AI Contract Is Signed
- The Agent Registry Nobody Maintains
- An Agent Has Gone Wrong. What Do You Stop?
- The Model Changed. Does the Approval Still Hold?
- A Demo Is Easy to Approve. A Service Is Harder.
- Where Should an Agent’s Tools Live?
Areas of research
AI agent authorization
These studies examine when an AI agent should be allowed to act, held for approval, or denied permission. They compare model decisions against explicit policies, measuring errors, response time, and reported certainty.
Algorithmic performance
Williams’ Law proposes a framework for understanding performance gains from algorithmic innovation. The research plan explains the proposal, reviews the available evidence, and sets out experiments needed to test it.
Publications, newest first
- Allow, Hold, or Deny: Comparing the Speed, Accuracy, and Reported Certainty of Four AI Models: A 300-case comparison of CLM, JEV, KEV, and GLM 5.3 FlashX examines accuracy, response time, and reported certainty when authorizing agent actions. The results show tradeoffs across the tested configurations, with no general winner.
- Published: 2026-09-27
- Audience: AI researchers, agent developers, and AI governance teams.
- Jev at the Agent Authorization Boundary: Evaluating TypeSafe’s Decision Model on Allow, Hold, and Deny: Two paired evaluations compare Jev with GPT-4.1 mini and DeepSeek V4.1 Flash on 270 constructed cases. The study examines false allows, withheld authorized actions, and refusals of actions that could proceed with human approval.
- Published: 2026-09-25
- Audience: Agent developers, security architects, and teams designing approval controls.
- Think Smarter, Not Harder: Algorithmic Innovation as the Key to Exponential AI Performance: This preprint examines algorithmic efficiency through Chain of Draft prompting and a mathematical framework relating AI performance to hardware capacity and algorithmic innovation.
- Published: 2025-03-02
- Audience: Platform teams, CIO/CTO offices, and applied AI leaders.
- Defining Williams' Law: The Power of Algorithmic Innovation: Performance gains come from disciplined algorithmic innovation, not only larger models.
- Published: 2025-02-28
- Audience: Enterprise architects, AI strategy leaders, and transformation teams.
- IT/OT Collaboration in Securing Canada's Energy Sector: Operational guidance for aligning strategy and standards with control implementation.
- Published: 2025-02-24
- Audience: Public-sector security leaders, energy operators, and governance teams.