AI SecurityBeginnerThreat2 validated evidence records

Shadow AI

Also known as: Unsanctioned AI, Unmanaged AI use

30 sec

Use of AI applications, agents, models, browser extensions, or integrations outside approved governance, visibility, or security controls.

Know

What is Shadow AI?

Shadow AI includes employees or teams adopting AI tools without security review as well as unsanctioned agents, plugins, accounts, models, or data flows. The security problem is not the mere existence of an unapproved tool; it is that sensitive data, identities, integrations, retention, and automated actions may sit outside normal controls and monitoring.

Why it matters

Verizon’s 2026 DBIR reported employee use of unapproved shadow AI tripled to 45%, increasing data-leakage risk. Mandiant likewise identifies shadow AI and lack of AI asset visibility as critical governance and security gaps.

Evidence, not hype

Validated in the real world

Every record is labeled by evidence type and source strength so an incident, a standard, and emerging research are never presented as if they are the same thing.

Technical ValidationMeasured outcome

2026 DBIR: vulnerability exploitation became the leading breach entry point

2026-06Verizon BusinessCross-sector

Verizon's 2026 DBIR overview reports exploitation of software vulnerabilities at 31% of breach entry points, third-party involvement at 48%, and employee use of unapproved shadow AI at 45%, alongside increasing AI-driven attack speed.

Why this is evidence

The DBIR provides broad breach-data evidence that vulnerability exploitation, third-party trust, and unmanaged AI use are not niche concerns in 2026; they are major enterprise exposure patterns.

See the source — Verizon: Vulnerability exploitation top breach entry point, 2026 DBIR finds
Emerging / ResearchResearch / emerging practice

Mandiant described AI moving from experimentation into operational adversary tradecraft

2026-03-09Google Cloud / MandiantCross-sector

Mandiant's 2026 AI Risk and Resilience report describes attackers moving beyond basic LLM use into adaptive code rewriting and agent-like workflows, while warning that shadow AI and poor AI asset visibility create significant enterprise risk.

Why this is evidence

This distinguishes two different 2026 AI security problems: adversaries using AI to improve attacks, and organizations creating unmanaged attack surface through rapid AI adoption.

See the source — Google Cloud / Mandiant: AI Risk and Resilience

Understand the mechanics

How it works

  1. 1

    A user or team adopts an AI tool outside approved procurement or IT workflows.

  2. 2

    Business data, code, credentials, files, or customer information is supplied to the tool.

  3. 3

    The tool may retain, process, train on, or route data in ways the organization cannot see.

  4. 4

    Plugins, agents, or integrations may gain access to additional systems.

  5. 5

    Security teams discover the exposure only after an incident, audit, or data-loss event.

Practice

What to watch for

  • Unapproved AI domains or browser extensions
  • Corporate data pasted into consumer AI accounts
  • Unknown AI OAuth integrations
  • API keys for unsanctioned models
  • Agents operating without registered owners
  • AI spend outside approved contracts

Perform

What to do

  1. 1

    Identify affected tools, users, data, and integrations.

  2. 2

    Determine whether sensitive information was retained or externally exposed.

  3. 3

    Revoke unauthorized OAuth grants, extensions, tokens, or agents.

  4. 4

    Move legitimate use cases into governed alternatives rather than relying only on prohibition.

How to reduce the risk

  • AI acceptable-use policy
  • Approved tool catalog
  • AI asset inventory
  • DLP and browser controls
  • OAuth governance
  • Secure enterprise AI alternatives
  • Employee education
  • AI vendor assessment

Business impact

  • Data leakage
  • Intellectual-property exposure
  • Compliance risk
  • Unknown retention
  • Unmonitored integrations
  • Loss of governance

What different roles should do

Employee

  • Use approved AI tools for company information

Security / GRC

  • Make sanctioned AI easy to use and maintain an inventory of AI applications and agents

Framework & standards context

  • NIST AI RMF
  • NIST CSF 2.0 Govern

Keep learning

Source transparency

Authoritative sources

Last reviewed: 2026-09-02