Prohibits using confidential information to train artificial intelligence models, restricts data scraping, and prevents automated extraction of protected data.
High Complexity
What This Clause Does
An AI usage restrictions clause explicitly prohibits the Receiving Party from using Confidential Information to train, develop, fine-tune, or improve artificial intelligence systems, machine learning models, or similar technologies. This includes restrictions on data scraping, automated extraction, and feeding information into generative AI tools that may retain or learn from the input.
Why This Clause Matters
Data Permanence: Once confidential information is used to train an AI model, it effectively becomes embedded in the model's weights and cannot be "deleted" or retrieved, making breaches irreversible.
Competitive Advantage: Training data is a core competitive asset. Preventing its use in AI training protects your proprietary methodologies, customer insights, and trade secrets.
Third-Party Exposure: Many AI tools are cloud-based services operated by third parties. Using confidential information with these tools may constitute unauthorized disclosure.
Regulatory Compliance: AI training with personal data may trigger GDPR, CCPA, or other privacy regulations, creating legal exposure for both parties.
Model Output Leakage: AI models can inadvertently reproduce training data in their outputs, potentially exposing confidential information to other users.
Legal Context
AI usage restrictions are a relatively new addition to NDA clauses, emerging primarily after 2022 with the widespread adoption of large language models. Courts have not yet extensively interpreted these provisions, making precise drafting essential. The clause should clearly define what constitutes "AI" or "machine learning" to avoid ambiguity, and should address both direct training and indirect uses (such as uploading to AI-powered analysis tools). Some jurisdictions may view overly broad AI restrictions as unenforceable restraints on technology use, so the restriction should be reasonably tailored to legitimate confidentiality concerns.
AI Training Prohibition
The Receiving Party shall not use any Confidential Information to train, develop, or improve any artificial intelligence system, machine learning model, or similar automated technology, whether owned by the Receiving Party or a third party.
Basic Version: Simple prohibition on AI training. Suitable for low-risk disclosures where the primary concern is preventing model training, but may not cover all AI-related uses.
Artificial Intelligence and Automated Processing Restrictions
The Receiving Party agrees to the following restrictions regarding artificial intelligence and automated data processing:
(a) No AI Training. The Receiving Party shall not use, or permit any third party to use, any Confidential Information to train, fine-tune, validate, test, develop, or improve any artificial intelligence system, machine learning model, neural network, large language model, or similar automated technology.
(b) No Automated Extraction. The Receiving Party shall not use any automated tool, bot, scraper, or similar technology to extract, compile, aggregate, or process Confidential Information, except with prior written consent from the Disclosing Party.
(c) AI-Assisted Analysis. If the Receiving Party wishes to use AI-powered tools to analyze Confidential Information, it must first obtain written approval from the Disclosing Party and ensure that such tools do not retain, learn from, or transmit the Confidential Information to any third party.
(d) Cloud AI Services. The Receiving Party shall not upload, input, or otherwise provide Confidential Information to any third-party AI service, chatbot, or cloud-based machine learning platform without prior written consent.
These restrictions shall survive termination or expiration of this Agreement for as long as the Confidential Information remains confidential.
Standard Version: Comprehensive coverage of AI training, data extraction, and third-party AI services. Appropriate for most business relationships involving sensitive data.
Comprehensive Artificial Intelligence Restrictions
1. Absolute Prohibition on AI Use. The Receiving Party absolutely and unconditionally agrees that it shall not, under any circumstances:
(a) Use any Confidential Information as input, training data, validation data, test data, fine-tuning data, or in any other capacity for any artificial intelligence, machine learning, deep learning, neural network, large language model, generative AI, or any current or future automated learning technology;
(b) Input, upload, paste, type, or otherwise provide any Confidential Information to any AI system, chatbot, virtual assistant, or automated analysis tool, whether or not such system claims to maintain confidentiality;
(c) Use any automated tool, script, bot, crawler, scraper, or programmatic method to access, extract, copy, compile, or process any Confidential Information;
(d) Permit any employee, contractor, agent, or affiliate to take any action prohibited by this section;
(e) Create any derivative dataset, embedding, vector representation, or other transformation of Confidential Information for AI-related purposes.
2. Certification Requirement. Upon request, the Receiving Party shall certify in writing that it has not used Confidential Information in connection with any AI system and shall provide reasonable documentation supporting such certification.
3. Audit Rights. The Disclosing Party shall have the right, upon reasonable notice, to audit the Receiving Party's AI systems and data practices to verify compliance with this section. The Receiving Party shall cooperate fully with any such audit.
4. Immediate Notification. The Receiving Party shall immediately notify the Disclosing Party if it discovers or suspects that any Confidential Information has been used in connection with AI training or automated processing, whether by the Receiving Party or any third party.
5. Remedies. Any violation of this section shall constitute a material breach entitling the Disclosing Party to immediate injunctive relief without the requirement of posting bond, in addition to all other remedies available at law or in equity. The Receiving Party acknowledges that damages for AI-related misuse would be difficult to calculate and agrees that the Disclosing Party may seek statutory damages of $500,000 per violation as liquidated damages.
6. Perpetual Duration. The restrictions in this section shall survive in perpetuity and shall not expire or terminate for any reason, recognizing that AI training creates permanent and irreversible incorporation of information.
Warning - Highly Restrictive: This version includes audit rights, certification requirements, and significant liquidated damages. The perpetual duration and $500,000 per-violation damages may be challenged as unreasonable. Receiving Parties should negotiate for more balanced terms.
1
Define AI Precisely: Push for a clear definition of what constitutes "artificial intelligence" or "machine learning." Overly broad definitions could inadvertently prohibit ordinary software tools like spell-checkers or search functions that use ML.
2
Carve Out Internal Tools: If your organization uses AI-powered internal tools (document management, search, etc.), negotiate an exception for these tools, provided they meet specified security requirements and do not transmit data externally.
3
Address Incidental Processing: Clarify whether the restriction applies to incidental AI processing (e.g., email providers using ML for spam filtering). Most parties will agree such background processing should be permitted.
4
Negotiate Approval Process: Instead of a blanket prohibition, propose a pre-approval process for specific AI uses. This gives the Disclosing Party control while allowing beneficial AI-assisted analysis when appropriate.
5
Resist Audit Rights: Audit rights for AI systems can be extremely intrusive and may expose your proprietary technology. If you must accept audits, limit them to documentation review rather than system access, require reasonable notice, and cap frequency.
6
Challenge Liquidated Damages: High per-violation damages may be unenforceable as penalties. Push for actual damages or reasonable liquidated damages tied to demonstrable harm, not arbitrary large amounts.
Perpetual Restrictions
AI restrictions that last "in perpetuity" or "forever" may be unenforceable and are impractical. Technology changes rapidly, and perpetual restrictions may prohibit future legitimate uses. Push for reasonable time limits.
Undefined "AI" Terminology
Clauses that prohibit "AI use" without defining the term could be interpreted to prohibit common software features. Ensure clear definitions that distinguish between prohibited AI training and permitted automation.
Unrestricted Audit Rights
Audit rights allowing the Disclosing Party to inspect your AI systems could expose your own trade secrets and proprietary technology. Require confidentiality protections and limit scope to compliance verification only.
Strict Liability for Third-Party AI
Language making you strictly liable if any third-party vendor uses AI on the data (even without your knowledge) creates unreasonable risk. Negotiate for liability only when you authorize or are negligent in preventing such use.
Excessive Liquidated Damages
Per-violation damages of $500,000 or more may be challenged as unenforceable penalties. Courts generally require liquidated damages to reasonably approximate actual harm, which is difficult to establish for AI training violations.