National AI Challenge 2026 · Team Incidentally Wonderful

The first-hour deputy for SME cyber incidents.

BreachDeputy helps an SME or MSP turn a suspected Microsoft 365 business-email-compromise incident into a safer, evidence-led professional hand-off. It preserves what happened, shows what matters, and keeps humans in control when the clock is running.

AI: bounded reasoning Authority: no write path to your systems Output: ready-to-review hand-off
Made for the judging criteria

One urgent problem. One visible, responsible outcome.

The first hour after a suspected payment-diversion or mailbox-compromise incident is a race against fragmented evidence, delayed escalation, and unsafe guesswork. BreachDeputy gives small organisations a structured path before specialist support arrives.

ProblemA high-pressure SME response gap
SolutionEvidence to hand-off in one flow
AIUseful reasoning, constrained authority
CommercialA repeatable MSP-led route to market
The product story

From a suspicious email to a professional starting point.

A deliberately narrow, fixture-led demonstrator for suspected Microsoft 365 business email compromise and payment diversion—not a claim of autonomous containment.

01

Problem

Small businesses can face a payment-diversion or mailbox-compromise incident without a security team or a rehearsed response. Important evidence, decisions, and escalation steps are easily scattered under pressure.

02

Solution

BreachDeputy guides the first hour from supplied email and diagnostic exports to an evidence record, a source-linked timeline, safe human tasks, and a professional hand-off pack.

03

How we use AI agents

Bounded model nodes extract, correlate, explain, ask targeted questions, label uncertainty, and draft. Deterministic controls keep policy, permissions, approvals, and consequential actions outside the model.

04

Monetisation

Our beachhead is the MSP: a repeatable first-hour workflow that can be offered across SME customers. Insurers, accountants, and incident responders are logical validation, distribution, and hand-off partners.

05

USP

We do not give the agent write access to your systems or your evidence. BreachDeputy’s advantage is controlled usefulness: provenance-led evidence, explicit uncertainty, human approval, and a hand-off that works even with upload/export-first inputs.

06

Lessons learned

In high-stakes cyber response, more autonomy is not always better. The product earns trust by making its boundaries visible—especially when attacker-controlled evidence tries to influence the AI.

The BreachDeputy difference

“The agent helps people understand and prepare. People retain the authority to act.”

✓
Evidence firstPreserve and organise supplied artefacts before an incident becomes harder to reconstruct.
✓
Safe by structurePrompt injection is treated as untrusted evidence—not as an instruction—and the model's only write tools are scoped to its own generated hand-off pack, never to the systems or evidence under investigation.
✓
Built for a real hand-offKeep facts, inferences, unknowns, human approvals, and next steps distinct for the professional who receives the case.
Team Incidentally Wonderful

Deep security, AI, data, and deployment experience.

BreachDeputy is designed by practitioners who understand both the incident-response reality and the limits of AI automation.

Michelle Simpson

Project lead · IR playbooks

Security leader with 20+ years’ experience, CISSP, Head of Pen Testing at Vertical Structure, and founder of Cyber Anthem. Brings incident-response, governance, and notification expertise.

LinkedIn ↗

Andrew Bolster

AI / ML · Agent architecture

PhD in Autonomous Systems and Senior R&D Manager in Data Science at Black Duck. Builds trusted, reproducible AI and decision-support systems.

LinkedIn ↗

Jordan Dickson

Infrastructure · DevOps

Senior DevOps Engineer at SAP with 6+ years of mission-critical systems experience. Leads dependable prototype deployment and operational engineering.

LinkedIn ↗

Claire Dickson

Security data · Evidence

Senior Security Data Engineer at Elastic and Python instructor. Brings log analysis, evidence capture, data engineering, and practitioner-community insight.

LinkedIn ↗
Contact

Interested in a safer first hour?

We are looking to learn from MSPs, incident responders, cyber insurers, accountants, and small-business operators who have seen the reality of a suspected business email compromise incident.

This demonstration form is intentionally front-end only. No personal information is sent or stored through this website.