Enterprise AI systems, including Microsoft Copilot and artificially intelligent (AI)-powered tools across Microsoft 365 and Azure, are often evaluated based on what they are designed to do. Far less attention is given to how they fail once deployed into real organisational environments.

This session introduces a different lens: failure is not random, rather it is predictable.

Drawing on criminological and social science frameworks, this session explores how harm and misuse emerge when systems interact with human behaviour, organisational dynamics, and incentives. Well-established criminological theories demonstrate that harmful outcomes are not anomalies, but the result of predictable conditions: motivated actors, accessible systems, and insufficient safeguards.

This session will be interactive. Using real-world examples of AI failures (including misuse, hallucination-driven errors, and manipulation of AI outputs) this session will allow attendees to examine how enterprise AI systems are shaped by their environments, not just their design.

Attendees will also learn how to anticipate where AI systems are most likely to fail within business workflows, identify behavioural and organisational risk factors, and have the opportunity to apply a structured way of thinking about risk that goes beyond technical performance, in a hypothetical case study exercise.

This session is designed to provide a new, practical perspective on AI risk – one that complements existing governance and security approaches, and helps organisations better prepare for the realities of deploying AI at scale.

Assumed Knowledge:

You should already have: A working understanding of how AI tools (such as Copilot or generative AI systems) are used within enterprise workflows; familiarity with concepts such as automation, data access, and user interaction within digital systems; basic awareness of AI risks such as hallucinations, bias, or misinformation This session will build on this knowledge and focus on deeper, system-level and behavioural risk analysis.

Practical Takeaways:

Understand how to identify predictable failure points in enterprise AI systems based on user behaviour, access patterns, and organisational context; learn how criminological concepts such as “opportunity structures” and “misuse incentives” apply directly to AI systems in business environments and how to ask better questions when evaluating AI deployments

Out of Session Scope:

We will NOT provide detailed technical deep dives into machine learning models or system architecture OR cover full organisational governance frameworks or compliance roadmaps OR focus on theoretical AI ethics