Congress shields AI developers from lawsuits over foreseeable errors

What it does

This bill grants artificial intelligence developers immunity from lawsuits when their AI systems make errors, provided the developers publicly release technical documentation (model cards and specifications) describing the AI's capabilities, limitations, and known failure modes, and keep that documentation updated within 30 days of deploying new versions or discovering material failures. The immunity applies only when the AI is used by licensed professionals (doctors, lawyers, etc.) serving clients, and does not shield developers from fraud claims or reckless/willful misconduct.

Why we flagged it

The operative mechanism is a conditional immunity grant to AI developers, framed as promoting 'responsible innovation' but functioning as a liability carve-out that exempts developers from negligence claims in exchange for transparency disclosures. The bill's core purpose is to reduce developer legal exposure, not to protect the public.

What the text implies

• Immunity applies even when a developer knows an AI system has a failure mode but fails to update documentation within 30 days—the 30-day window creates a grace period during which developers can deploy known-defective systems without liability risk.
• Professionals (doctors, lawyers) remain liable to their clients for AI errors, even when the error stems from a developer's failure to disclose known limitations. Liability shifts from developer to professional, creating perverse incentives for professionals to avoid AI use or to over-rely on developer disclosures they cannot independently verify.
• Trade secret redactions in model specifications are permitted if the developer provides 'written justification'—no independent review mechanism exists, allowing developers to withhold safety-critical information while claiming trade secret protection.
• The bill applies only to 'learned professionals' in licensed fields, leaving consumers and non-professional users (e.g., small businesses using AI for hiring, lending, or content moderation) without any immunity framework—creating a two-tier system where professional users get some recourse but ordinary people do not.
• Fraud and 'knowing misrepresentation' are carved out from immunity, but the bill does not define what constitutes knowing misrepresentation of AI capabilities—developers could argue that incomplete or outdated model cards do not rise to fraud if they were technically accurate at the time of release.

Section numbers refer to the bill text the analysis read — linked under Primary records below.

Who it affects

Ordinary citizens and professionals lose the ability to sue AI developers for foreseeable errors, even when those errors cause real harm in critical domains (healthcare, law, finance). While transparency requirements are valuable, they do not compensate for the loss of legal recourse—a professional harmed by a known failure mode the developer failed to update cannot recover damages. The immunity is broad enough to shield developers from negligence claims, shifting liability risk onto professiona

Who stands to gain

• AI developers and large language model companies
• AI product vendors and distributors
• Professional services firms deploying AI (healthcare, legal, financial services)

Named in the bill

AI developers, Learned professionals, National Artificial Intelligence Initiative Act of 2020, Professional services sectors (healthcare, law, finance)

Where it stands

• Jun 12, 2025 — Introduced · Congress.gov: “Introduced in Senate”
• Jun 12, 2025 — Referred to Senate Committee on Commerce, Science, and Transportation · Congress.gov: “Read twice and referred to the Committee on Commerce, Science, and Transportation”

Dates and quoted wording are Congress.gov's action record; the timeline shows status changes, not every procedural step.

Money around this bill

3 lobbying clients named this bill on 3 disclosure filings across 2 quarters, Dec 2025 to Jun 2026. Those filings disclosed $6,340,000 in lobbying spend. A filing names 28 bills on average, so that figure is what each filing reported, not a share belonging to this bill.

More lobbying clients named this bill than 61% of bills with at least one filing.

Cynthia Lummis, the sponsor, reported $558,192 in PAC receipts in the 2026 cycle.

• Meta Platforms, Inc. and Various Subsidiaries — $5,990,000 on 1 filing
• Business Software Alliance — $320,000 on 1 filing
• Bsa the Software Alliance (formerly Bsa Business Software Alliance Inc) — $30,000 on 1 filing

Lobbying Disclosure Act filings through Jul 20, 2026. A filing shows who paid to lobby on a bill it names, not what changed.