Event Programme

AMLUCS 2026: A Preview of What's Coming

The submissions are in and our reviews are complete! We’ve been overwhelmed by the volume and quality of submissions, and the diverse spread of content from across government, industry and academia.

Please see below an early peek at the conference content. The full programme and line-up will be published soon. We look forward to seeing you at UK's leading conference for the practical application of AI and cybersecurity.

Keynotes

Hyrum Anderson is cofounder and CEO of Abundant Security.  Previously, he was Senior Director of AI & Security at Cisco, via an acquisition of Robust Intelligence where he served as CTO.  Much of his career has been focused on defence and security, having directed research projects at MIT Lincoln Laboratory, Sandia National Laboratories, Mandiant, as Chief Scientist at Endgame (acquired by Elastic), and Principal Architect of Trustworthy Machine Learning at Microsoft.  Hyrum co-founded the Conference on Applied Machine Learning in Information Security (CAMLIS), one of AMLUCS’ the main inspirations.

Second keynote - to be announced!

Panel Discussions

Where Do We Start? A Practical Guide for Organisations Adopting AI-Driven Cybersecurity

With AI adoption accelerating across security operations, many organisations are still wrestling with where to begin. This panel brings together operational leaders, researchers and technology practitioners to discuss practical adoption strategies, success stories, common pitfalls, governance considerations and how to identify use cases that provide measurable value.

Secure Enough? How Regulation Shapes AI Security

As governments and regulators increasingly focus on AI risk and accountability, security teams must understand how regulation impacts both system design and deployment. This panel will explore emerging regulatory approaches, standards, assurance expectations and how organisations can balance innovation with compliance.

Hands-On Workshop

The Agent's Inner Monologue: Catching Compromise in the Reasoning Trace (Secure Agentics)

Most agent defences operate at the edges: input filters, guardrails, output classifiers and system prompts. But what if the most valuable security signal lies inside the agent itself? Inspired by recent advances in monitoring agent reasoning processes, this highly practical workshop explores how observing an agent's reasoning trace can reveal attacks and behaviours that action-level monitoring alone may miss. Gain hands-on experience using Adrian, an open-source agent runtime monitoring platform.

Training

This two-day course teaches a working AI red teaming methodology you can apply to production AI systems on Monday morning. Across ten modules, attendees move from the foundations of AI red teaming through hands-on attack execution, multi-turn techniques, methodology, and reporting, finishing with how the evidence produced supports ETSI EN 304 223 conformity assessments [link].

Technical Presentations/Posters

AI Security Threats, Vulnerabilities & Mitigations

  • Agentic Large Language Models for Autonomous Cyber Operations (BAE Systems Digital Intelligence)

  • Cross-Layer Detection for Scaled AI Misuse (Microsoft AI Safety Platform)

  • Detecting Attacks Against AI Systems (University of Oxford)

  • Do Model Representations Know When a Jailbreak Is Working? (HiveTrace)

  • Hessian Spectral Signatures for Backdoor Detection (Alan Turing Institute)

  • Latent Backdoors (Loughborough University)

  • Packets as Binary Images: Per-Packet Intrusion Detection for OT (BMT)

  • Per-Agent Compliant, Collectively Unsafe: A Deployable Collective Bound for Colluding AI Agents (Omega)

  • Planning for Quantum Backed AI (Deep Cyber Ltd)

  • Reproducible Agentic Environment System - RAES (Palo Alto)

  • Security of Novel Compute in the AI Stack (DSTL)

  • The Agentic Attack Surface: Latent Failures in Autonomous Workflows (Queens University Belfast)

  • Thematic Review and Gap Analysis of AI Security (Lancaster University)

  • Two Graders for One Agent: Deterministic Rule Verification and LLM-as-a-Judge in an Adversarial Agent Evaluation Environment (Alice)

Agent Capabilities, Offensive & Defensive Cyber Operations

  • Addressing the Root Cause: Automated Fuzzing Harness Generator for Patch Completeness Testing  (The Alan Turing Institute, University of Cambridge)

  • Agentic Security Benchmarks are Executable Threat Models: Measuring Risk Relocation in AI Defences (Advai)

  • Auditing and Recovering Agentic Cyber Defence Systems (Cardiff University)

  • Automated Experimentation in Cyber Labs (BT)

  • Capture the Flag, Miss the Malware: Hijacking Agentic Code Scanners with Planted Rewards (Alice)

  • From Detection to Response (Riskaware)

  • Identifiable Causal Discovery with Latent Time-varying Confounder Estimation (LaTICE) (Loughborough University)

  • Investigating Agentic Cybercrime (Deep Cyber Ltd)

  • Measured Feasibility of Agentic Cyber Defence at the Land Tactical Edge (Aeris-UK)

  • Novel Compute for Data-Efficient Reinforcement Learning in Autonomous Cyber Defence (Frazer Nash Consultancy)

  • Project AIX: Deploying Cyber Resilience Agents on a Real Network (NCSC)

  • Sustainable Cybersecurity in the Era of Hyperconnected Agentic AI and Cyber-Physical Systems (Palo Alto)

  • Reasoning, Not Rationalising: Mechanistic Guardrails for Agentic AI in Critical National Infrastructure (Tikos Technologies)

  • Reproducing Mythos Findings with Public LLMs and Disclosing Two New Linux Kernel Zero-Days (Vidoc Security Lab)