Risk · R-150 · Intellectual Property & Legal Liability
Confidential information disclosed to AI tools
Trade secrets, legally privileged material or information held in confidence for others are entered into AI tools and leave the organisation's control.
- Severity
- High
- Audience
- Board · Heads-of · Practitioner
- Frameworks
- Intellectual Property · LLM02 · Chapter 6
- Responsible AI domain
- Lifecycle stage
- Reviewed
- 2026-11-11
What it is
Staff use AI tools on the material in front of them: source code, product plans, draft contracts, board papers, another company's documents received under a confidentiality agreement. Entering it into an external tool is a disclosure. The legal consequences differ from those for personal data, covered in personal data in prompts and context. A trade secret is protected only while reasonable steps are taken to keep it secret. Legal privilege can be lost by disclosure. A confidentiality agreement may forbid passing information to any third party, which a tool's provider is.
How it happens
- An engineer pastes proprietary code into a public assistant to find a fault.
- A lawyer uses a general tool to summarise privileged advice.
- Documents received in a transaction are uploaded for analysis.
- A meeting assistant records and transcribes a confidential discussion by default.
- The approved tool is less capable, so staff use another.
Why it matters
Protection that depends on secrecy may be lost for good. The organisation may be in breach of contracts with others and of professional duties. If the provider trains on inputs, the information may later surface in answers to other users, as vendor reuse of organisation data describes.
Warning signs
- No guidance tells staff what classes of information may go into which tools.
- Confidentiality agreements have not been reviewed for their effect on AI use.
- Transcription and note-taking tools join meetings automatically.
- Network data shows large uploads to AI providers.
Who owns it
The owner of the AI policy is accountable for the rules. Information owners classify what they hold. Legal advises on privilege and third-party confidentiality. Security monitors and blocks.
In practice
During due diligence on an acquisition, an analyst uploads the target's customer contracts to a public tool to extract key terms. The confidentiality agreement with the target prohibits disclosure to anyone outside the deal team. The same pattern generalises to any confidential material handled by people with a deadline and a capable tool one tab away.
Mitigating controls
Primary controls
- AI acceptable use policy Directive
- Defines what may be entered into which tools. A policy approved by top management states what AI use is permitted, what is prohibited, what every system must pass through, who is accountable, and where to take a case the policy does not settle.
- Prompt and context data loss prevention Preventive
- Blocks or redacts restricted information in prompts and uploads. Prompts, uploads and retrieved context are inspected before they reach a model, and blocked, redacted or flagged where they contain personal or restricted data.
Supporting controls
- AI contract terms Directive
- Imposes confidentiality and no-training terms on providers. Contracts with AI suppliers cover data use, change notification, information and audit rights, sub-contracting, incident notification, liability and exit.
- Shadow AI discovery Detective
- Finds unapproved tools where the policy is bypassed. Network, procurement, expense and software records are reviewed on a set cycle to find AI tools and features in use that are absent from the inventory.
Framework and clause references
| Framework | Clause | Title |
|---|---|---|
| NIST Generative AI Profile (NIST AI 600-1) | Intellectual Property | Intellectual Property Names exposure of trade secrets. |
| OWASP Top 10 for LLM Applications | LLM02 | Sensitive Information Disclosure Sensitive information disclosure. |
| SS2/21: Outsourcing and Third-Party Risk Management | Chapter 6 | Outsourcing agreements Confidentiality terms with providers. |
Sources
- SS2/21: Outsourcing and Third-Party Risk Management Bank of England PRA · official text
- NIST Generative AI Profile NIST · framework
- OWASP Top 10 for LLM Applications OWASP GenAI Security Project · framework
Related terms
How to cite
Axiom Verity (2026). Confidential information disclosed to AI tools. Reference Library, R-150. https://axiom-verity.com/risks/confidential-information-disclosed-to-ai-tools. Reviewed 2026-11-11.
Found an error or a better source? Report a correction.