Connecticut: AI Ethics Guidance for Law Firms

Bottom line for a 5-50 attorney Connecticut firm: Connecticut now has a binding state-court rule, and it arrived as a Practice Book amendment rather than an ethics opinion. Effective June 23, 2026, Section 4-9 requires anyone who uses generative AI in creating or editing a Superior Court filing to independently verify all citations, legal authorities, and evidence it produced, and failure “may result in court-imposed sanctions, including, without limitation, the entry of a nonsuit or default judgment.” Revised Section 4-2(b) makes signing the filing a certification of compliance, and since July 14, 2026 the appellate rules (Sections 60-4, 62-6(d), and 85-2) carry the same certification and a listed sanction into Supreme Court and Appellate Court filings. On July 31, 2026, in TOV Realty, LLC v. Suarez, the Supreme Court imposed its first sanctions for AI-hallucinated citations, under Rule 1.1 rather than the new rules. There is still no Connecticut bar ethics opinion, so ABA Formal Opinion 512 continues to fill the ethics-side gap alongside Rules 1.1, 1.4, 1.6, 3.3, 5.1, 5.3, and 8.4.

Start with: a written AI use policy, a citation verification log that can support the Section 4-2(b) certification on Superior Court filings, and vendor due-diligence records for each AI tool in use.

Practice Book Section 4-9: the operative rule

Citation: Conn. Practice Book Sec. 4-9, “Generative Artificial Intelligence (‘Generative AI’) Compliance,” adopted by the judges of the Superior Court on June 11, 2026 and published in the Connecticut Law Journal on June 23, 2026, effective on publication. Primary source PDF (rule text at pp. 14-16).

Section 4-9(b) is the operative sentence: “any person who uses generative AI in the creation or editing of any document filed with the court shall independently verify all citations, legal authorities or evidence produced by generative AI.” Failure to do so “may result in court-imposed sanctions, including, without limitation, the entry of a nonsuit or default judgment.”

Four features distinguish it from the disclosure-style orders other states adopted.

It reaches any person, not just attorneys, so self-represented parties are covered on the same terms. It imposes no disclosure duty: nothing requires a filer to say that AI was used, only to verify what it produced. Section 4-9(d) places responsibility “solely” on the filer, foreclosing an argument that a vendor’s or staff member’s error shifts the burden. And Section 4-9(a) enumerates specific tools by name, including ChatGPT, Co-Counsel, Harvey, Ironclad, DeepSeek, and Grok.

Section 4-2(b) is the enforcement hook: signing a filing now certifies compliance with Section 4-9, which folds the verification duty into the signature block rather than leaving it as a free-floating obligation.

Note the effective date carefully. Most amendments in the same Law Journal notice take effect January 1, 2027; Sections 4-2 and 4-9 took effect on publication, June 23, 2026.

Appellate filings. The appellate courts followed three weeks later. Amendments to Sections 60-4, 62-6, and 85-2 of the Rules of Appellate Procedure, approved by the Appellate Court on June 23 and by the Supreme Court on June 30, 2026, took effect on publication in the Connecticut Law Journal on July 14, 2026 (primary source PDF). Section 60-4 gives generative AI the same meaning as in Section 4-9. Section 62-6(d) makes the signature on any appellate document a certificate that the signer “has complied with the requirements of Section 4-7 regarding personal identifying information and Section 4-9 regarding generative AI.” Section 85-2 adds “[f]ailure to independently verify all citations and legal authorities produced by generative AI in compliance with Sections 4-9 and 62-6 (d)” to the actions subject to appellate sanctions, which include a filing bar for a definite period, a fine under General Statutes § 51-84, and costs and attorney’s fees to the opposing party. Sections 4-2 and 4-9 themselves remain Superior Court rules; the appellate rules import the duty rather than extend the section.

No operative bar opinion (pending activity)

No formal or informal opinion on attorney AI use has issued from the Connecticut Bar Association Committee on Professional Ethics. To monitor, watch the CBA’s informal opinions index. An active standing committee, the CBA’s Generative AI Committee, runs CLE programming but has published no written guidance as of 2026-09-27, and the informal opinions index runs through Informal Opinion 25-04 with nothing on AI. Within the Judicial Branch, the Committee on Artificial Intelligence in the Connecticut Legal System (CAIC) has been evaluating whether the Practice Book and Rules of Professional Conduct require amendment for AI.

The ethics side and the rules side have moved at different speeds, and the distinction matters. There is still no Connecticut bar opinion, which is why this page’s guidance status remains pending. But the Practice Book was amended: Section 4-9 took effect 2026-06-23 and is binding on filings, as described above. A firm looking for its Connecticut obligations should work from the court rule, not wait on the bar.

In the absence of Connecticut-specific guidance, ABA Formal Opinion 512 (Generative Artificial Intelligence Tools, 2024-07-29) is the most directly applicable framework. Connecticut adopted ABA Model Rule 1.1 Comment 8 (technology competence) effective 2014-01-01, and Connecticut’s RPCs parallel the ABA Model Rules covered by Op 512 in all relevant respects.

What federal courts in Connecticut require for AI use in filings

On 2025-09-12, the U.S. District Court for the District of Connecticut issued a Notice to Counsel and Litigants Regarding AI-Assisted Research. Per the notice, use of AI without verification of the accuracy of its output implicates Fed. R. Civ. P. 11. The notice announces a no-tolerance policy for hallucinated legal propositions or other severe misstatements of law: “Such filings will often result in sanctions absent reasonable excuse.” Affirmative pre-filing disclosure of AI use is not required. The notice cautions that attorneys and pro se litigants “should exercise great caution in submitting any AI-generated language in filings before the Court,” and that “[u]se of AI without verification of the accuracy of the information it generates” implicates Rule 11. It applies district-wide to attorneys and pro se litigants.

What Connecticut state courts require for AI use in filings

Connecticut adopted one. Practice Book Section 4-9 took effect June 23, 2026 and is described in full above. It requires independent verification of every citation, legal authority, and item of evidence produced by generative AI in any Superior Court filing, imposes no disclosure duty, reaches self-represented parties as well as attorneys, and is enforced through the signature certification in revised Section 4-2(b). Sanctions reach as far as nonsuit or default judgment. For Supreme Court and Appellate Court filings, Sections 60-4, 62-6(d), and 85-2 of the appellate rules, effective July 14, 2026, impose the same verification certification and list its breach as a sanctionable act.

This supersedes the earlier position on this page. The 2026 Practice Book as published on 2026-01-01 contained no AI-specific rule, and a certification rule considered by the Superior Court Rules Committee had been deferred; the June 2026 amendment is what changed. Firms should still monitor the Connecticut Law Journal Practice Book index for further amendments.

Separately, the Connecticut Judicial Branch’s Responsible AI Policy Framework, Version 2.1 (2024-02-01) governs court staff and judicial-branch AI procurement, not attorneys filing in Connecticut courts.

Has anyone been sanctioned for AI use in Connecticut?

Yes. In TOV Realty, LLC v. Suarez, SC 21183, and Kosel Equity, LLC v. MacGregor, SC 21184 (Conn. July 31, 2026) (per curiam), the Supreme Court confronted “for the first time a lawyer’s use of generative artificial intelligence” resulting in hallucinated citations in filings before it. Primary source PDF. Attorney Ian G. Gottlieb of GLG Law, LLC filed applications for certification to bring public interest appeals under General Statutes § 52-265a in two cases, then briefs in each; together they contained approximately seven erroneous and unverified citations. He had researched and Shepardized the drafts in LexisNexis, then pasted them into ChatGPT to improve the writing, and the tool added or altered case citations without changing the substantive propositions, so neither he nor the firm colleagues who reviewed the documents re-verified the cites. Amici represented by the Jerome N. Frank Legal Services Organization alerted the court. At a show-cause hearing on July 7, 2026, Gottlieb admitted through counsel that the filings violated the duty of competence under Rule 1.1.

The court ordered Gottlieb to complete six hours of continuing legal education in ethics and law office management, three of them on generative AI, beyond the twelve-hour annual minimum; ordered Gottlieb and the firm each to donate $1,000 to the CT Bar Institute, Inc.; and required a compliance report within six months with a courtesy copy to the Statewide Grievance Committee, which the court said “is not a referral for further disciplinary action.” A footnote assigns the firm a share of the responsibility for having had no policies on the responsible use of generative AI in place. The sanction rests on Rule 1.1, not on Section 4-9, which post-dates the filings; the court closed by noting the Superior Court judges’ new rule. This matter was the direct catalyst for Section 4-9.

What AI-related rules are pending in Connecticut?

Superior Court Rules Committee AI citation certification rule: adopted. At its 2025-02-09 meeting, the Rules Committee considered requiring attorneys using AI for legal research to certify independent verification of all citations, with case-ending sanctions available. No action was taken at the February 2026 meeting and the matter was deferred.

It did not stay deferred. The judges of the Superior Court adopted Practice Book Section 4-9 on 2026-06-11, effective on publication 2026-06-23, and it carries exactly the certification-plus-sanctions structure the Committee had been weighing: independent verification of the citations, legal authorities, and evidence generative AI produced, enforced through the signature under revised Section 4-2(b), with nonsuit or default judgment available. Verification was always required under Rules 3.3 and 8.4; Section 4-9 turns it into an explicit, sanctions-bearing record of noncompliance. This item is closed, and the operative text is described at the top of this page.

CBA Generative AI Committee and CAIC. Both bodies are studying attorney-facing AI questions but have produced no published written guidance.

Public Act 26-15 (SB 5, 2026), “An Act Concerning Online Safety.” The House passed the bill on 2026-05-01 and the Governor signed it on 2026-05-27 (bill status; act text). The omnibus act does not directly regulate attorney conduct. Its provision of most potential relevance to a firm is on automated employment-related decision technology (sections 7 to 12, effective 2026-10-01): a deployer that, on or after 2027-10-01, uses such technology to generate output for making, or as a substantial factor in making, an employment decision about an employee or applicant in Connecticut must first give that person written notice of the technology, its purpose, the categories and sources of personal data it analyzes, and contact information, and must disclose AI interactions with employees or applicants unless obvious. Violations are unfair trade practices enforced solely by the Attorney General. Section 3 (effective 2027-07-01) directs the Commissioner of Economic and Community Development to plan an AI regulatory sandbox, with recommendations due by 2028-01-01. A firm that uses AI tools in hiring or personnel decisions may wish to review them against the 2027-10-01 notice requirements.

Connecticut SB 2 (2025). Passed the Senate 32-4 on 2025-05-15 but did not receive a House floor vote and died at end of session. Its deployer obligations are not Connecticut law.

Public Act 25-113 (signed 2025-06-25). The act’s amendments to the Connecticut Data Privacy Act took effect 2026-07-01 (act text). Section 9 requires a controller’s privacy notice to include “a statement disclosing whether the controller collects, uses or sells personal data for the purpose of training large language models.” Section 6 also broadened who the CTDPA covers. As amended, it applies to persons that conduct business in Connecticut or target Connecticut residents and that, in the preceding calendar year, controlled or processed the personal data of at least 35,000 consumers (down from 100,000), or that control or process consumers’ sensitive data, or that offer consumers’ personal data for sale, in each case subject to the statute’s exemptions. Because the sensitive-data prong has no volume threshold, a firm handling client health, financial, or other sensitive information may wish to confirm with counsel whether it, or a vendor processing that data, is now within the CTDPA.

Connecticut AG AI Advisory (2026-02-25). Attorney General William Tong issued a memorandum clarifying that existing Connecticut civil rights, consumer protection (CUTPA), privacy (CTDPA), and antitrust laws apply to AI-driven conduct without new legislation. Firms using AI in client-intake screening, marketing, or automated billing should review those workflows against the civil rights and consumer protection categories.

How do malpractice carriers in Connecticut treat AI use?

No Connecticut-specific malpractice carrier guidance on AI has been publicly issued. Standard lawyers professional liability policies were not written with generative AI in mind. A broker has reported a competitor attaching a manuscript AI exclusion to a non-standard lawyers program; the endorsement itself is not public. Whether standard policy language reaches AI-related malpractice claims is unresolved, and some cyber-policy structures cap AI-adjacent claims well below overall policy limits. Connecticut firms using AI in client matters should confirm in writing with their LPL carrier or broker whether the current policy covers claims arising from AI-generated work product. Document what exclusions or sublimits apply and what documentation the carrier expects.

What does my Connecticut malpractice carrier ask about AI at renewal?

No Connecticut malpractice carrier has published AI-specific application items in materials reviewed for this entry. The substantive standard a carrier or broker will reference is ABA Formal Opinion 512 plus the Connecticut RPCs, with the D. Conn. notice as the operative federal-filing standard. Plan on producing: a written AI use policy, attorney and staff training records, vendor due-diligence records, a citation-verification protocol, and supervision records for AI-generated work product. If Public Act 25-113 reaches the firm or its vendors, the firm’s privacy notice must reflect the LLM training data disclosure by 2026-07-01. Confirm application items directly with the carrier or broker.

What documentation should a Connecticut firm keep on file?

Month one (foundational)

1. (Owner: managing partner + firm administrator) Written AI use policy (Rules 1.1, 1.6, 5.1, 5.3; ABA Op 512). Enumerate approved tools, prohibit submitting unverified AI output to courts or clients, address which tools may receive client-identifying information, and assign verification responsibility per matter.
2. (Owner: litigation lead) Citation verification log satisfying Practice Book Section 4-9 (Sec. 4-9; Rules 3.3, 8.4; Fed. R. Civ. P. 11). Section 4-9 scopes its duty to material “produced by generative AI,” so what the rule compels is independent verification of the citations, legal authorities, and evidence the tool produced, not of every item in a filing that happened to involve AI. Document that verification against primary sources before submission. For Superior Court filings this is no longer prudential: Section 4-9 requires it and Section 4-2(b) makes the signature a certification that it was done. A firm may of course adopt a broader verify-everything policy, and Rules 3.3 and 8.4 already reach human-produced errors, but that is a firm choice rather than what Section 4-9 imposes.
3. (Owner: litigation lead) D. Conn. and Superior Court filing checklist (Sec. 4-9; Rule 3.3; Fed. R. Civ. P. 11). Review the D. Conn. notice before each federal filing. For Superior Court filings, work from the adopted text of Section 4-9, effective 2026-06-23, rather than from the earlier proposed certification standard. Sections 4-2 and 4-9 were adopted by the judges of the Superior Court and govern Superior Court filings; Supreme Court and Appellate Court filings are governed by the parallel appellate amendments (Sections 60-4, 62-6(d), and 85-2, effective July 14, 2026), under which the signature on an appellate document certifies Section 4-9 compliance and a verification failure is a listed ground for appellate sanctions. Note that Section 4-9 reaches self-represented parties and imposes no disclosure duty, so a template built around disclosing AI use solves the wrong problem.

Months two and three (operational documentation)

1. (Owner: firm administrator + outside IT) Vendor due-diligence records (Rule 1.6). For each AI tool used on client matters: documented review of data handling, retention, training-data use, and security certifications; data processing agreement where available; dated review updated when terms change.
2. (Owner: firm administrator) Attorney and staff training records (Rules 1.1, 5.1, 5.3). Date, attendees, and content of training on AI tool limitations, hallucination risk, confidentiality obligations under Rule 1.6, and the firm’s AI policy. Update when tools or guidance change.
3. (Owner: managing partner) Supervision protocol (Rules 5.1, 5.3). Document how managing attorneys supervise AI use by lawyers and non-lawyers to the same standard as other work product.

Months four to six (per-matter discipline)

These are recurring practices, not one-time projects: each new matter exercises them.

1. (Owner: matter lead attorney) Per-matter AI use notes (Rules 1.1, 3.3). Brief attorney note on any matter where AI generated research, drafted pleadings, or produced filed content: which tool, what tasks, how output was verified.
2. (Owner: managing partner + billing partner) Engagement letter and disclosure decision log (Rules 1.4, 1.5). No Connecticut per se disclosure requirement exists today. Document the disclosure analysis for each matter where AI materially affects scope, cost, or confidentiality. Address AI tool use in retainer terms where matter sensitivity warrants.
3. (Owner: managing partner) CTDPA and carrier inquiry review (Rule 1.6). Confirm whether the firm or any vendor is covered by the CTDPA as amended by Public Act 25-113 (in effect since 2026-07-01, including the sensitive-data coverage prong). Document the written conversation with the LPL carrier or broker on AI coverage and required documentation.

We are building practitioner resources for firms working through this list. The newsletter covers new resources as they ship.