Court Rules on AI-Generated Filings: What to Disclose Where

Standing orders and practice directions on AI-generated filings — the US certification orders, Canadian court declarations, and a compliance workflow.

Do Courts Require Disclosure of AI Use? The Current Answer

Some do, most don't — and the ones that do impose their requirements through judge-specific standing orders and court practice directions rather than rules of civil procedure, which means the duty varies not just by jurisdiction but by courtroom. In the United States, the trend began in May 2023 when Judge Brantley Starr (N.D. Tex.) issued a standing order requiring every filer to certify either that no portion of a filing was drafted by generative AI or that any AI-drafted language was verified by a human using traditional legal databases. Dozens of federal and state judges followed with their own variants; other courts and several bar groups pushed back, arguing existing Rule 11 obligations already cover it — the Fifth Circuit, notably, proposed and then declined to adopt a circuit-wide certification rule after public comment.

In Canada, courts moved at the institutional level. The Court of King's Bench of Manitoba issued the first practice direction (June 2023) requiring parties to disclose when and how AI was used in preparing materials; the Supreme Court of Yukon followed within days. The Federal Court of Canada's notice to the profession (December 2023, since updated) requires a declaration when generative AI created content in documents submitted to the Court and urges human verification. Alberta's courts issued a joint notice emphasizing caution and verification, and other superior courts have published guidance ranging from disclosure requirements to reminders that counsel remain fully responsible for filings. Provincial practice continues to evolve — check the current notices for every court you file in.

This article maps the landscape as part of our AI in Legal Practice library; the sanctions cases that motivated these rules are detailed in AI hallucinations in legal research.

What the Orders Actually Require: Three Models

Reading the orders together, three regulatory models emerge:

• Certification model (many US standing orders): file a certificate stating either that generative AI was not used, or that all AI-generated content was verified for accuracy by a human against authoritative sources. Judge Starr's order is the archetype; variants differ on whether research-only use triggers the duty.
• Disclosure model (Manitoba, Yukon, Federal Court of Canada): declare when AI was used to generate content in court documents — typically a statement in or accompanying the filing identifying that AI-generated material is included. The Federal Court frames it as a declaration for content "created or generated by AI."
• Verification-reminder model (Alberta's joint notice, many state courts, and the default everywhere else): no filing-by-filing disclosure, but explicit warnings that counsel bear full responsibility for accuracy and that fabricated authority will be treated as misconduct — i.e., Rule 11, FRCP 26(g), and their Canadian analogues do the work.

Appellate and administrative tribunals are writing their own variants, and self-represented litigants complicate the picture further: several Canadian practice directions apply to all parties precisely because unrepresented filers have produced a disproportionate share of hallucinated authority, and US courts have sanctioned or admonished self-represented parties on the same basis. For counsel, that means AI-fabricated authority may now arrive from the other side of the file — checking the opposing party's citations through a citator has quietly become part of competent motion practice, and courts appreciate counsel who flag phantom authority early rather than letting a judge discover it.

Scope questions matter in practice. Most orders target generative AI creating content, not spell-checkers, translation aids, or research databases with AI features — but definitions vary, and a few early orders were broad enough to arguably capture Westlaw's AI summaries. Where an order is ambiguous, the safe reading is the broad one, disclosed in a sentence; no judge has ever sanctioned a lawyer for over-disclosing.

Underlying Duties: Why the Baseline Never Changed

Whatever the local disclosure regime, four baseline duties govern every AI-assisted filing everywhere. FRCP 11(b) (and provincial rules with the same effect) makes signing a filing a certification that legal contentions are warranted by existing law — a fabricated case fails that certification regardless of any AI disclosure. The duty of candor (ABA Model Rule 3.3; FLSC Model Code 5.1-2) prohibits knowingly making false statements to a tribunal, and courts have read reckless non-verification as functionally equivalent once a lawyer knows a generative tool was used. Competence requires understanding hallucination before relying on the tool. And supervision rules put partners on the hook for associates' AI-assisted filings — several sanctions decisions have named the reviewing lawyer, not just the drafter.

Note also what the orders do not do: none of them prohibit AI use, and judges who issued the strictest certification requirements have said explicitly that the tools are permissible with verification. The orders regulate process, not technology — a distinction that matters when advising risk-averse partners, because the compliant answer to a standing order is never "we don't use AI" (an assertion that is almost certainly false somewhere in the workflow) but "we verify and we can prove it." Courts have likewise declined to treat AI assistance itself as diminishing the weight of submissions; what they punish is the transfer of the lawyer's verification duty to a machine that cannot bear it.

The disclosure orders, in other words, are procedural belt-and-suspenders on substantive duties that already existed — which is why compliance is best treated as a verification-workflow problem rather than a form-filing problem.

One coordination problem deserves attention in multi-jurisdiction practices: the same brief may be filed in a certification-model courtroom, a disclosure-model court, and three courts with no requirement at all. Firms handle this with a highest-common-denominator rule — run the full verification protocol and keep the AI-use log on every matter, then generate whatever paper each court requires from that single record. The alternative, tailoring diligence to each court's minimum, guarantees that eventually a filing crosses jurisdictions faster than the workflow does.

A Compliance Workflow for Litigators

The workflow that keeps firms clean across every model:

• Check before every filing: the assigned judge's standing orders (US federal), local rules, and — in Canada — the court's current practice directions and notices to the profession. Build it into the filing checklist beside page limits.
• Maintain an AI-use log per matter: which tools touched which documents and for what. When a certification or declaration is required, the log makes it a two-minute task rather than an archaeology project.
• Verify per protocol: every authority pulled in a real database, every quotation checked, citator run on everything — the protocol from our hallucinations guide, executed and initialed before signature.
• Standard certification language: keep house templates for both certification-model and disclosure-model courts so compliance never depends on drafting under deadline pressure.
• Correct fast if something slips through: the sanctions record is unambiguous that prompt, candid correction dramatically mitigates outcomes, while defensiveness aggravates them.

Firms should also watch the trajectory: standing-order proliferation has slowed as courts conclude existing rules suffice, but the disclosure regimes that exist are sticky, and new appellate guidance arrives regularly — the regulatory context is surveyed in our companion piece on bar and law society AI rules. Meanwhile the same courts' future litigants are asking ChatGPT which lawyer to hire; our ChatGPT for law firms hub and AI SEO service cover that front, and you can book a free strategy call for a firm-wide plan.

Put AI to Work in Your Firm — the Right Way

LexScale.ai helps law firms across Canada and the United States adopt AI for growth — from client-facing intake and content systems to the visibility that puts your firm inside AI answers.

Frequently Asked Questions

This article is general information, not legal or ethics advice. Professional-conduct rules on AI are evolving and vary by jurisdiction — always verify current requirements with your state bar, law society, or regulator before adopting any AI workflow.

Related Articles

Billing Ethics for AI-Assisted Legal Work  ·  AI Contract Drafting for Lawyers: What Works  ·  AI Document Review & Discovery: From TAR to LLMs  ·  AI Hallucinations in Legal Research: Risks & Fixes  ·  Bar Rules on AI for Lawyers: US & Canada Guidance  ·  Can Lawyers Use ChatGPT? Ethics, Risks & Workflow