Vendors are Deflecting Liability for Anything 'AI'

Large vendors are rewriting terms to deflect liability for AI; smaller vendors are following suit. See exactly what they're deflecting.

Vendors are putting the AI risk on you

We alert you when they do.

✓Disclaimed liability for AI features

✓Indemnity clauses that name you

✓Rights claimed over your prompts

✓Access-by-agent restrictions

Major vendors such as JPMorgan and Nasdaq are amending their terms of service to deflect liability for AI risks. Smaller vendors are following suit, including but not limited to Docker, ElevenLabs, Splunk, LexisNexis, and Predictive Index.

New clauses disclaim responsibility for:

Actions and output from AI features and integrated third-party AI
External agents using vendors' systems
and the confidentiality of data processed by AI.

Liability for first-party AI features and integrated third-party AI

Vendors shipping AI features and integrating with external AI capabilities are increasingly publishing provisions that deflect responsibility for harmful AI outputs. This includes agents taking erroneous actions, inaccurate outputs in sensitive contexts, and IP-infringing outputs.

Examples include:

ElevenLabs disclaims liability for any output generated by third-party providers; they use OpenAI, Anthropic, and Gemini models as a major part of the service.
LexisNexis disclaims responsibility for harms resulting from users relying on web-sourced content pulled in by its AI, including with respect to IP-infringing outputs.
Splunk disclaims risks from its own agentic AI except in cases of gross negligence, defects, or misconduct.
Docker goes the extra mile, covering damages from their AI capabilities and use of third-party AI, even going so far as to require customers to indemnify Docker in court should harm occur.

Liability for third-party AI using their services

In addition to vendors deflecting risk for use of their AI features and AI integrations, vendors are also seeking to disclaim any risks from their features being used by external AI tools. Examples include JPMorgan Chase and, more recently, Robinhood (which has announced Agentic Trading via MCP).

Some of these terms are very difficult or impossible to adhere to. For example, JPMorgan's Terms of Use require an AI Agent to identify itself, stating:

"AI Agents must promptly and truthfully respond to any question or prompt seeking to determine if interactions are coming from a human or an AI Agent."

This is fundamentally impossible for non-deterministic LLMs to achieve with 100% efficacy.

Their terms go on to indicate that users are fully responsible for all actions taken by agents they leverage, even "if it performs unintended actions or exceeds the authority you gave it."

Liability for customer data privacy and handling

The third group of changes concerns the privacy of model inputs and associated outputs. These changes provide vendors the rights to reuse or share inputs and outputs from AI features, stating they are not liable for this data being divulged. As examples:

Nasdaq reserves "the right to reproduce and/or otherwise use customer provided information in any manner for purposes of training artificial intelligence technologies," and disclaims responsibility for the accuracy of its GenAI output.

Predictive Index's addendum, effective May 15, 2026, warns that prompts "may be divulged or revealed by the AI Functionality to someone else," may be retained to "influence future outputs," and that "complete isolation or deletion of a Prompt from the AI Functionality may not be technically feasible."

Real Examples from Vendors

Chase

Terms of Use

Expanded unauthorized access to cover AI agents “designed to circumvent” access controls, and set conditions for permitted agent access: agents must identify themselves, “clearly and deterministically label all agentic traffic,” and answer truthfully when asked if they are human. Chase can “limit, prevent, suspend, or terminate” agent access at its discretion, and holds the deployer “responsible for any actions or errors” an agent causes — even if it “exceeds the authority you gave it.”

Docker

Subscription Service Agreement

Disclaims all liability for “any action taken, Output generated, or omission by an AI Agent,” and applies that “regardless of whether such actions were authorized, anticipated, or foreseeable.” An expanded indemnity makes the customer “defend and hold Docker … harmless” against third-party claims from agent actions. Docker also “disclaims any responsibility and liability for Outputs from Third-Party Product providers.”

ElevenLabs

Image & Video Terms

Disclaims liability for anything tied to an outside provider: “ElevenLabs shall have no liability arising from or relating to any Third-Party Provider,” including any “Output generated or provided by such Third-Party Provider.”

Splunk

Specific Terms for Splunk Offerings

An Agentic AI Features section places primary responsibility on organizations to review, approve, and ensure lawful use of all AI-generated actions, including human-in-the-loop controls. Splunk appears to be only liable for defects in its AI features, its gross negligence or misconduct, or breach of its explicit obligations; organizations appear to be responsible for indemnifying Splunk for claims arising from misuse, misconfiguration, or failure to apply required oversight and human-in-the-loop controls.

Nasdaq

Dorsey Wright Terms of Use

Reserves “the right to reproduce and/or otherwise use customer provided information in any manner for purposes of training artificial intelligence technologies.” Disclaims responsibility for GenAI output: “It is your responsibility to verify the accuracy and non-infringement of GenAI-generated content before using it.”

LexisNexis

General Terms and Conditions

Disclaims liability for AI output drawn from the open web: “LN disclaims all liability for any damage, loss, or harm arising from your reliance on Web-Sourced Content.” It also does not warrant that such content is free of third-party intellectual property rights.

Predictive Index

Client Software Agreement (AI Addendum)

New addendum warns that prompts “may be divulged or revealed by the AI Functionality to someone else,” may be retained and used to “influence future outputs,” and that “complete isolation or deletion of a Prompt from the AI Functionality may not be technically feasible.” The customer grants a broad license to use prompts and outputs; PI disclaims liability for client confidential information the AI reveals if it was previously submitted by the client in a prompt.

Stay up to date when more vendors make this change

The table above is a sample. We track every vendor quietly rewriting its terms to offload AI risk onto customers.