Agentic AI, Shadow AI, and Supply Chain Risk Form One Enterprise Control Gap
What happened
The open-access survey Below the Waterline: A Survey of Agentic AI Security, Shadow AI Governance, and LLM Supply-Chain Risk in the Enterprise was published October 21, 2026 in Applied Sciences. It maps the intersection of three enterprise AI risk categories that most organizations manage independently. The paper argues that agentic AI deployments, unauthorized employee AI use (shadow AI), and third-party model supply chain risks share common control failure points. These include incomplete asset inventories, weak vendor due diligence, and poor identity controls for AI systems. Incident response programs were also not designed for AI-specific failure modes. The authors review existing frameworks including the NIST AI Risk Management Framework (AI RMF 1.0) and Playbook and the OWASP Top 10 for Large Language Model Applications. Controls addressing each risk area exist in isolation but are rarely integrated into a single governance posture. Organizations treating these as distinct programs will systematically underestimate their total exposure. This is especially true as regulators under the EU AI Act (Regulation (EU) 2024/1689) begin requesting cross-cutting evidence of risk management.
Why it matters
- ·Regulators are increasingly asking for unified AI risk evidence. The EU AI Act (Regulation (EU) 2024/1689) and sector supervisors expect organizations to demonstrate integrated oversight across AI systems, not separate programs for each risk type. A siloed governance structure will produce audit gaps that enforcement reviews will find.
- ·Shadow AI is not just a policy problem; it is a supply chain entry point. When employees use unapproved AI tools, those tools often connect to external model providers whose provenance, training data, and security controls the enterprise has never reviewed. This links unauthorized use directly to third-party and data-exposure risk.
- ·Agentic AI amplifies every gap in the inventory and identity controls that shadow AI and supply chain risk already expose. An agent may operate under a shared or unknown identity and pull from an unvetted model provider. It may also act outside approved boundaries. This represents a compounded failure across three control domains at once.
Stay ahead of stories like this
Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.
Recent issues
- Three separate incidents this week each created a new compliance action item: verify your AI vendors' data sourcing, demand agent monitoring coverage guarantees, and audit trusted app identities in your cloud environment.29 Oct
- Regulators, litigants, and attackers are all moving faster than most organizations' AI governance frameworks, shifting leverage away from deployers and toward those who can prove control, accountability, and honest disclosure.21 Oct
Free every Thursday. Unsubscribe anytime.
Governance controls affected
What to do now
- ☐Ask your security and compliance teams whether your AI asset inventory covers not only approved AI tools but also AI agents, employee-installed tools, and the external model providers those tools connect to.
- ☐Map your existing vendor due diligence process to confirm it reaches the upstream model providers behind the AI tools your employees and systems use, not just the immediate vendor relationship.
- ☐Review whether your AI agents operate under their own named, logged identities or under shared user credentials, and escalate to your engineering team if agents cannot be distinguished from human users in audit logs.
- ☐Confirm that your AI incident response playbook covers failure modes specific to agents (for example, an agent acting beyond its permitted scope or a shadow AI tool leaking data), not only conventional software incidents.
- ☐Assign a named owner to consolidate the oversight of agentic AI, shadow AI, and AI vendor supply chain risk into a single risk register, and set a date for the first integrated review.
What to watch next
Supervisory pressure to produce unified AI risk evidence is rising across jurisdictions. The EU AI Act (Regulation (EU) 2024/1689) enforcement cycle is advancing. Sector regulators in financial services and healthcare are beginning to probe whether organizations can demonstrate integrated oversight rather than point controls. Teams should watch for further guidance from the EU AI Office on cross-cutting risk evidence. They should also monitor whether the Five Eyes Guidance on the Careful Adoption of Agentic AI Services and similar advisory outputs are cited in formal supervisory communications.