“The strongest cybersecurity strategy begins with a simple question: Where else does this information exist?” – MJ Martin
The Invisible Copies That Could Become Your Greatest Cybersecurity Risk
Artificial intelligence is rapidly transforming Canadian businesses, governments, municipalities, healthcare organizations, and utilities. From drafting reports to analyzing engineering drawings and searching corporate knowledge bases, AI has become an everyday productivity tool. Yet beneath this remarkable progress lies a growing problem that many executives have never heard of. It is called shadow data.
The recent IronCore Labs AI Shadow Data White Paper argues that organizations often focus on protecting their primary databases while overlooking the many secondary copies of sensitive information created as AI systems operate. Those overlooked copies may ultimately become more attractive targets for cybercriminals than the original databases themselves.
Canada should pay particular attention to this issue. Municipal utilities, provincial governments, healthcare networks, engineering firms, and critical infrastructure operators increasingly depend upon AI to improve efficiency while simultaneously managing highly sensitive information protected under legislation such as PIPEDA and various provincial privacy laws.
Every Photocopy Is Another Risk
Imagine a municipality owns the master blueprint for its water distribution system. The original drawing is carefully stored inside a secure vault with alarms, cameras, and restricted access.
Now imagine dozens of photocopies are quietly left on desks, inside filing cabinets, in meeting rooms, and in contractors’ trucks.
Which document is most likely to disappear?
The original remains heavily protected. The forgotten copies become the weakest links.
Shadow data follows the same principle. The official customer database may be well secured, but AI applications often create temporary or permanent copies while processing prompts, generating summaries, indexing documents, or building semantic search databases. These copies frequently exist outside traditional governance processes.
Why AI Needs More Copies Than Traditional Software
Traditional software generally retrieves exactly what it needs from a database.
Large Language Models work differently.
Before AI can answer questions intelligently, it often converts documents into mathematical representations known as vector embeddings. These vectors allow the system to search for concepts rather than exact words, dramatically improving the quality of responses through Retrieval Augmented Generation, commonly called RAG.
A useful analogy is a library catalogue.
Instead of searching every book on every question, the catalogue first identifies the shelves most likely to contain the answer. AI performs a similar function using vectors. The challenge is that the catalogue itself may contain enough information to reveal the contents of the books if inadequately protected.
Recent academic research has demonstrated that vector embeddings may leak portions of their original source information through inversion and inference attacks, making them an emerging cybersecurity concern recognized by the 2025 OWASP Top 10 for LLM Applications.
Why This Matters for Canadian Utilities
Consider a Canadian water utility deploying AI to help engineers answer operational questions.
The AI may access GIS records, SCADA documentation, maintenance reports, leak histories, engineering drawings, customer billing records, and asset inventories.
The utility’s primary databases may remain secure. However, AI may generate searchable indexes, conversation histories, prompt logs, cached documents, and vector databases containing fragments of all that information. If those repositories receive less protection than the operational systems, attackers may choose the easier path.
This represents a shift in cybersecurity thinking. Organizations must now inventory not only where their authoritative data resides but also where AI silently replicates portions of it.
Governance Must Evolve Alongside AI
Canada’s AI adoption strategy should extend beyond selecting the right model. It must include governance over prompt logging, data retention, encryption, access controls, lifecycle management, and continuous monitoring of AI-generated repositories.
The IronCore paper highlights application layer encryption and stronger protection of vector databases as emerging approaches for reducing exposure, while recognizing that AI architectures introduce entirely new attack surfaces requiring equally innovative safeguards.
Ultimately, organizations should treat AI not simply as another software application but as a new participant in the enterprise that continuously reads, summarizes, indexes, and occasionally duplicates information.
Summary
Artificial intelligence promises extraordinary productivity gains across Canada. Yet every technological revolution creates new responsibilities.
The greatest cybersecurity threats are often not the data we deliberately protect but the copies we never realized existed. Shadow data reminds us that information security is no longer about guarding a single vault. It is about understanding every path that information travels after leaving that vault.
As Canadian municipalities, utilities, and private enterprises embrace AI, they should ask an important strategic question. Have we secured only our original data, or have we also secured the shadows that AI leaves behind?
References
IronCore Labs. AI Shadow Data White Paper, March 3, 2025.
file_8B1BE6BE-E48B-4B51-B54F-D2D1180A7947.pdf
OWASP Foundation. OWASP Top 10 for LLM Applications 2025.
“Text Embeddings Reveal (Almost) As Much As Text,” cited in the IronCore Labs white paper.
file_8B1BE6BE-E48B-4B51-B54F-D2D1180A7947.pdf
About the Author:
Michael Martin is the Vice President of Technology with Metercor Inc., a Smart Meter, IoT, and Smart City systems integrator based in Canada. He has more than 40 years of experience in systems design for applications that use broadband networks, optical fibre, wireless, and digital communications technologies. He is a business and technology consultant. He was a senior executive consultant for 15 years with IBM, where he worked in the GBS Global Center of Competency for Energy and Utilities and the GTS Global Center of Excellence for Energy and Utilities. He is a founding partner and President of MICAN Communications and before that was President of Comlink Systems Limited and Ensat Broadcast Services, Inc., both divisions of Cygnal Technologies Corporation (CYN: TSX).
Martin served on the Board of Directors for TeraGo Inc (TGO: TSX) and on the Board of Directors for Avante Logixx Inc. (XX: TSX.V). He has served as a Member, SCC ISO-IEC JTC 1/SC-41 – Internet of Things and related technologies, ISO – International Organization for Standardization, and as a member of the NIST SP 500-325 Fog Computing Conceptual Model, National Institute of Standards and Technology. He served on the Board of Governors of the University of Ontario Institute of Technology (UOIT) [now Ontario Tech University] and on the Board of Advisers of five different Colleges in Ontario – Centennial College, Humber College, George Brown College, Durham College, Ryerson Polytechnic University [now Toronto Metropolitan University]. For 16 years he served on the Board of the Society of Motion Picture and Television Engineers (SMPTE), Toronto Section.
He holds three master’s degrees – in business (MBA), communication (MA), and education (MEd). As well, he has three undergraduate diplomas and seven major certifications in business, computer programming, internetworking, project management, media, photography, and communication technology. He has completed over 80 next generation MOOC (Massive Open Online Courses) [aka Micro Learning] continuous education programs in a wide variety of topics, including: Economics, Python Programming, Internet of Things, Cloud, Artificial Intelligence and Cognitive systems, Blockchain, Agile, Power BI, Big Data, Design Thinking, Security, Indigenous Canada awareness, and more.
Martin in a volunteer, a photographer, a learner, a technologist, a philosophizer, and a romantic optimist.
Shadow data is an interesting problem. One concern I would have is in using AI to do a bid. If a vendor creates a response to an RFP using a free version of ChatGPT, can a competitor unknowingly leverage that work in its bid?