AI in arbitration

Summary#

This bill would add a new rule to the federal arbitration law to allow parties to use artificial intelligence (AI) in arbitration under certain conditions. It lists specific uses allowed, requires written consent by the parties who will use AI, bars vacating an award for errors caused by AI, and bans “generative” AI. The stated goal is to permit AI tools in arbitration while limiting some legal challenges based on AI mistakes.

• Main change: Inserts a new section permitting a party to use AI in arbitration for certain tasks, subject to written consent by the parties involved.
• Allowed uses: finding relevant law (including case law), recommending document order/form/relevance, filing supporting documents when facts are not in dispute and a genuine legal question exists, and delivering a final decision to the parties.
• Broad catch-all: allows “such other purpose as a party determines necessary.”
• Consent and risk rule: each party must provide written consent to use AI, and a party that uses or agrees to use AI may not seek to vacate or modify an award solely because of an error from that AI.
• Limitation: generative AI is explicitly prohibited in arbitration under this law.
• Unclear points: the bill does not define “artificial intelligence” or “generative artificial intelligence,” and it does not explain how use, disclosure, oversight, or responsibility will work in practice.

What it means for you#

• Parties in arbitration (businesses, consumers, employees, insurers):

  • You could use AI tools to help find law, organize documents, file certain documents, or to receive a final decision from AI — but only if all parties give written consent and the AI used is not “generative.”
  • If you agree to use AI, you cannot later seek to set aside or change an award solely because the AI made an error.
  • It is unclear whether a party can refuse to “use” AI while another party uses it; the consent language may require all parties to both consent and use the AI.

• Arbitrators and arbitration providers:

  • May need to handle cases where parties use AI for research, document work, filings, and possibly for delivering decisions.
  • Providers might have to set rules on what AI tools are allowed, but the bill does not prescribe specific standards or oversight.

• Lawyers and representatives:

  • Could use AI to prepare filings and research, subject to client consent and the generative AI ban.
  • Must consider the bar on vacating awards for AI errors when advising clients about risks.

• General public / courts:

  • Courts reviewing arbitration awards may be limited in undoing an award that results from an agreed AI use if the party relying on the AI consented to it and seeks vacatur only because of AI error.

• What is unclear:

  • No definition of “artificial intelligence” or “generative artificial intelligence.”
  • No rules about required disclosure to arbitrators, documentation of AI use, audits, or how to assess AI errors.
  • No effective date or transition rules are provided in the text shown.

Expenses#

No publicly available information.

• Possible costs that could follow from the bill (not stated in the bill):
  • Arbitration providers or courts may face administrative costs to adopt policies, technology, or training to handle AI use.
  • Parties or law firms could incur costs to purchase, certify, or run AI tools and to document their use.
  • There could be savings from faster research, document handling, or processing if AI speeds tasks, but the bill gives no estimate.
  • Enforcement or disputes about what constitutes “generative” AI or about consent could create legal costs.

Proponents' View#

• The bill appears intended to allow modern tools into arbitration to improve efficiency and legal research.
• Supporters may argue that permitting AI for specified tasks can speed case preparation and document handling.
• Allowing AI to deliver decisions or file documents (with consent) could reduce administrative delay in straightforward cases.
• The rule preventing vacatur for AI errors could be framed as creating certainty and limiting rounds of post-award litigation over technical tool mistakes.
• Banning generative AI might be presented as a safety measure to limit risks from AI that creates new text or content.

Opponents' View#

• One concern is the lack of definitions: the bill does not define “artificial intelligence” or “generative artificial intelligence,” leaving wide uncertainty about which tools are allowed or banned.
• The consent language is unclear and may be read to require every party both to consent and to use AI, which could be impractical or unfair.
• Allowing AI to “deliver a final decision” without clear rules raises questions about whether a human arbitrator must actually decide the case and how to ensure fairness and accountability.
• The bar on vacating awards for errors “solely” due to AI may limit remedies even when serious mistakes affect outcomes; it is unclear how mixed-error cases would be handled.
• The bill gives no rules for transparency, testing, recordkeeping, auditing, or review of AI outputs, which may make it hard to detect bias, errors, or misconduct.
• It does not address power imbalances where one party may feel pressured to consent to AI use to save costs or speed, potentially affecting fairness.