AI Drafted S-1s and the Illusion of Immunity

Cooley GO Public and the Duty to Verify an AI Drafted S-1

Executive Summary

Cooley launched GO Public with OpenAI on September 17, 2026, an artificial intelligence tool that drafts Form S-1 registration statements, and Section 11 liability still falls on every human signatory. GO Public runs on ChatGPT Enterprise and an agentic workflow built with OpenAI’s forward deployed engineers, combining a company’s own data with Cooley’s precedent library to produce an initial S-1 draft in minutes rather than days. The tool changes who does the typing. It does not change who answers for what the finished registration statement says to an investor who later loses money.

GO Public is the second release under Cooley AI, the firm’s dedicated artificial intelligence unit, following IPO GO, an interactive Form S-1 drafting guide the firm launched in February 2025. Cooley advised on 180 venture financings and public offerings worth 51.5 billion United States Dollars in 2025, the precedent library behind the new tool. OpenAI describes Cooley as a design partner in building the underlying agentic harness, with checkpoints at which a lawyer must review the output before the workflow advances.

Escott v. BarChris Construction Corp., 283 F. Supp. 643 (S.D.N.Y. 1968), still defines the reasonable investigation every signer, officer, and director must conduct before a registration statement becomes effective, and nothing in the Securities Act of 1933 excuses that investigation because a machine wrote the first draft. Newer questions remain genuinely open: whether the Securities and Exchange Commission will ask issuers to disclose that an AI tool helped draft a filing, whether malpractice insurers will require a documented verification protocol, and how a law firm protects a client’s confidential financial data inside a large language model’s working context. This report sets out the verification process a startup’s counsel should run on any AI generated S-1 draft before it reaches the printer.

Detailed Findings

What Cooley GO Public Actually Does

Cooley LLP announced GO Public on September 16, 2026, describing it as a proprietary offering built with OpenAI that uses ChatGPT Enterprise to help lawyers and management teams prepare a Form S-1. OpenAI’s account of the partnership names Cooley as a design partner that worked with OpenAI’s forward deployed engineers to build an agentic harness on top of ChatGPT Work, one that synthesizes client provided information, public filings, and Cooley’s own precedent library into a tailored starting document. The workflow includes defined checkpoints at which a lawyer must review or validate the output before the process continues to the next drafting step.

Cooley Chief Innovation Officer David Wang described the release as changing legal work by bringing intelligence to every step of the drafting process, while capital markets co-chair Dave Peinsipp framed the goal as freeing lawyer time for the judgment calls that shape a transaction rather than the mechanical first pass at a disclosure document. Law360 reported that the tool could shorten the drafting phase of an S-1 by as much as a month, though the full path from confidential submission to effectiveness still runs through the pace of review at the SEC’s Division of Corporation Finance. GO Public follows IPO GO, an interactive Form S-1 drafting guide Cooley launched on February 12, 2025 around a satirical sample filing, and it is the first release under Cooley AI, the firm’s dedicated artificial intelligence services unit, which the firm intends to extend into private company, fund formation, and merger practice.

Section 11 Imposes Liability No Matter Who Drafts the Filing

Section 11 of the Securities Act of 1933, codified at 15 U.S.C. § 77k, creates a nearly strict liability claim for any purchaser of a security when the registration statement, at the time it became effective, contained an untrue statement of material fact or omitted a material fact required to make the statements not misleading. Every person who signs the registration statement, every director of the issuer, and every underwriter named in it can be sued, and the purchaser does not have to prove reliance or even that the misstatement caused the loss. Section 12(a)(2) of the Securities Act, 15 U.S.C. § 77l(a)(2), imposes a parallel liability running through the prospectus itself, which means the exposure attaches to the finished document a company files, not to the process used to produce a draft of it.

The one affirmative defense available to everyone but the issuer is the due diligence defense in Section 11(b)(3): a defendant who conducted a reasonable investigation and had reasonable grounds to believe, and did believe, that the statements were true escapes liability. Escott v. BarChris Construction Corp., 283 F. Supp. 643 (S.D.N.Y. 1968), remains the controlling illustration of how demanding that standard is. The court found that outside directors, the underwriters, and the company’s own lawyer who signed the registration statement had not conducted a reasonable investigation because each relied on management’s representations rather than independently checking the facts against the company’s books, contracts, and operating records. Rule 176 under the Securities Act, 17 C.F.R. § 230.176, lists the type of issuer, the signer’s role, and the reasonable availability of the relevant facts among the circumstances a court weighs in judging reasonableness, and none of those factors changes because a large language model produced the sentence under review instead of a first year associate.

Certifications and Disclosure Controls Under the Securities Laws

Once a company goes public, Section 302 of the Sarbanes-Oxley Act of 2002, codified at 15 U.S.C. § 7241, requires the chief executive officer and chief financial officer to personally certify that periodic reports contain no material misstatement and that the company maintains disclosure controls reasonably designed to surface material information to them. The Securities and Exchange Commission’s 2020 amendments to Regulation S-K, adopted in Release No. 33-10825, require issuers to disclose the risks most significant to that specific company rather than generic boilerplate under Item 105, and the companion amendments in Release No. 33-10890 require a discussion under Item 303 of known trends and uncertainties reasonably likely to affect operating results.

A risk factor or trends discussion an AI tool assembles from precedent filings runs a particular risk under both requirements. Language written for a different company’s facts is, by definition, closer to the boilerplate the 2020 amendments were adopted to eliminate than to the company specific disclosure Item 105 now demands. A chief financial officer who signs a Section 302 certification without confirming that AI drafted trend language actually reflects the company’s own internal reporting has not satisfied that certification, regardless of how polished the sentence reads.

The Professional Responsibility Layer

The American Bar Association addressed a lawyer’s obligations when using a generative AI tool in Formal Opinion 512, issued July 29, 2024 and titled Generative Artificial Intelligence Tools. The opinion holds that Model Rule 1.1 competence requires a lawyer to understand a generative AI tool’s capabilities and limitations before relying on it, that Model Rule 1.6 confidentiality requires informed client consent before client information enters such a system, and that Model Rules 3.3 and 8.4(c) require independent verification of every citation, fact, and legal statement the tool produces before it goes into a filing.

Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), supplies the cautionary record behind that rule. A New York lawyer submitted a brief citing cases ChatGPT had invented, opposing counsel and the court could not locate the citations, and the presiding judge sanctioned the attorney for failing to verify the tool’s output before filing it. A Form S-1 carries no docket number, but the discipline is the same: a citation, a regulation, or a data point an AI drafting tool inserts into a registration statement needs the same independent check a securities lawyer would run on an associate’s first draft, because an SEC reviewer reads a registration statement for precisely this kind of error.

The SEC Is Already Testing AI Related Accuracy Claims

The Securities and Exchange Commission’s willingness to scrutinize AI related representations is not theoretical. On March 18, 2024, the SEC settled charges against Delphia (USA) Inc. for 225,000 United States Dollars after finding the Toronto based investment adviser could not substantiate claims that it used artificial intelligence to analyze client data and predict which companies were about to succeed. On the same day, the SEC settled a parallel action against Global Predictions Inc. for 175,000 United States Dollars over an unsubstantiated claim to be the first regulated AI financial adviser, together with disclaimer language the agency found created a misleading impression that clients had waived nonwaivable claims.

Neither case involved a Form S-1, but both establish that the agency will test an AI related representation for substantiation the same way it tests any other disclosure claim. That posture bears directly on two separate questions a startup’s counsel now faces: how the company describes its own use of artificial intelligence in the business section of its registration statement, and how a reviewer might eventually treat a disclosure document that the issuer’s own counsel used an AI tool to draft in the first place.

Building a Verification Protocol for an AI Drafted S-1

The verification process a startup’s counsel should run before filing a registration statement an AI tool helped draft is the process BarChris already requires, applied with the discipline that a tool with no memory of the underlying transaction demands. Every quantitative statement, from a revenue recognition policy to a customer count to a market size figure, needs to be traced to a specific document in the data room by a named individual, not accepted because it matches the tone of the precedent draft the tool was trained on. Every risk factor needs a session with the officer or department actually responsible for that risk, the same due diligence call the BarChris court punished the defendants for skipping.

Counsel should retain the prompts submitted and the drafts returned as part of the deal file. The Securities and Exchange Commission does not currently require this record, but a documented verification trail is the closest modern equivalent to the due diligence file that separated the BarChris defendants who escaped liability from the ones who did not. Before any company information enters GO Public or a comparable tool, the engagement letter or a separate written consent should address what Formal Opinion 512 requires: client authorization for the specific handling of confidential information the tool involves, not a general technology clause written before generative AI existed.

The officer and director due diligence sessions, the accountant’s comfort letter procedures, and the underwriter’s independent investigation should proceed exactly as they would for a fully human drafted registration statement. Section 11 assigns liability to the people who sign the document, not to the software that helped write the first version of it.

Want more industry insights? I publish daily analysis on the intersection of deep tech and venture law.

Subscribe to The Innovation Attorney on Substack

Open Questions

  1. Will the SEC’s Division of Corporation Finance begin asking issuers, through a comment letter or a new item of Regulation S-K, to disclose when an AI tool produced the initial draft of a registration statement. A new disclosure item would convert what is currently an internal drafting choice into a public fact investors and boards can weigh.
  2. Will professional liability insurers require a documented AI verification protocol as a condition of coverage for law firms and for the officers and directors of companies that use tools such as GO Public. Startups negotiating their own directors and officers coverage should watch this question closely, since a carrier’s underwriting standard could become a working verification requirement before any regulator writes one.
  3. How do Cooley and OpenAI contractually protect a client’s nonpublic financial and operational data once it is processed inside ChatGPT Enterprise during the drafting workflow. The public announcements describe the workflow but not the data handling terms, a gap that matters to any general counsel deciding whether to authorize the tool’s use under Rule 1.6.
  4. Will other firms named as early adopters of OpenAI’s legal platform, including Sullivan and Cromwell, converge on the same review checkpoints Cooley describes, or will competitive pressure to move faster erode the human review step OpenAI itself calls mandatory. The answer will shape whether the due diligence defense remains equally available across firms using different AI drafting tools.
  5. Could a hallucinated or overgeneralized risk factor that an AI tool inserted into an effective registration statement become the basis of a Section 11 claim distinct from the underlying business risk it was meant to describe. No court has yet tested that theory, and the first case to raise it will set the practical boundary of AI assisted drafting in public offerings.

Source List

  1. Escott v. BarChris Construction Corp., 283 F. Supp. 643 (S.D.N.Y. 1968). Decided 1968. The controlling federal decision applying the Section 11 due diligence defense fact by fact, relevant because it still sets the standard against which an AI drafted registration statement must be checked.
  2. Securities Act of 1933 § 11, 15 U.S.C. § 77k; § 12(a)(2), 15 U.S.C. § 77l(a)(2). Enacted 1933. The statutory liability provisions that attach to a registration statement and a prospectus regardless of how the document was drafted.
  3. Securities Act Rule 176, 17 C.F.R. § 230.176. Effective 1982. Lists the circumstances a court weighs in judging whether a defendant’s investigation of a registration statement was reasonable.
  4. Sarbanes-Oxley Act of 2002 § 302, 15 U.S.C. § 7241. Enacted July 30, 2002. Requires the chief executive officer and chief financial officer of a public company to certify the accuracy of periodic reports and the adequacy of disclosure controls.
  5. Securities and Exchange Commission, Modernization of Regulation S-K Items 101, 103, and 105, Release No. 33-10825 (Aug. 26, 2020). Requires company specific risk factor disclosure under Item 105 rather than generic boilerplate.
  6. Securities and Exchange Commission, Management’s Discussion and Analysis, Selected Financial Data, and Supplementary Financial Information, Release No. 33-10890 (Nov. 19, 2020). Modernizes Item 303 trends and uncertainties disclosure.
  7. American Bar Association, Formal Opinion 512, Generative Artificial Intelligence Tools (July 29, 2024). The controlling ethics guidance on a lawyer’s duty to verify, protect confidentiality, and disclose the use of generative AI tools.
  8. Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023). Decided 2023. The sanctions decision arising from fabricated AI generated case citations, illustrating the verification failure Formal Opinion 512 was written to prevent.
  9. Securities and Exchange Commission, In the Matter of Delphia (USA) Inc., Release No. IA-6539 (Mar. 18, 2024). A 225,000 United States Dollar settlement for unsubstantiated claims about artificial intelligence used to serve investors.
  10. Securities and Exchange Commission, In the Matter of Global Predictions Inc., Release No. IA-6540 (Mar. 18, 2024). A 175,000 United States Dollar settlement for an unsubstantiated artificial intelligence claim and misleading liability disclaimer language.
  11. Cooley LLP, Cooley Launches GO Public With OpenAI (Sept. 16, 2026). The primary announcement describing the tool, its OpenAI partnership, and the named executives quoted throughout this report.
  12. OpenAI, How Cooley Is Accelerating IPO Work With ChatGPT (Sept. 17, 2026). OpenAI’s own account of the technical architecture, the design partnership, and the human review checkpoints built into the workflow.
  13. Law360, Cooley Launches AI System To Speed IPO Document Drafting (Sept. 18, 2026). Independent legal trade press reporting on the projected drafting time savings and the limits of that savings on the overall IPO timeline.
  14. Cooley LLP, Cooley Launches IPO GO (Feb. 12, 2025). The announcement of GO Public’s predecessor tool, relevant to understanding Cooley’s multi year build toward an AI assisted drafting product.

Bibliography

  1. American Bar Association. Formal Opinion 512: Generative Artificial Intelligence Tools. Washington, D.C.: American Bar Association, July 29, 2024.
  2. Cooley LLP. “Cooley Launches GO Public With OpenAI.” Cooley.com, September 16, 2026.
  3. Cooley LLP. “Cooley Launches IPO GO.” Cooley.com, February 12, 2025.
  4. Escott v. BarChris Construction Corp., 283 F. Supp. 643 (S.D.N.Y. 1968).
  5. Law360. “Cooley Launches AI System To Speed IPO Document Drafting.” Law360 Pulse, September 18, 2026.
  6. Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023).
  7. OpenAI. “How Cooley Is Accelerating IPO Work With ChatGPT.” OpenAI.com, September 17, 2026.
  8. Sarbanes-Oxley Act of 2002 § 302, 15 U.S.C. § 7241.
  9. Securities Act of 1933 § 11, 15 U.S.C. § 77k.
  10. Securities Act of 1933 § 12(a)(2), 15 U.S.C. § 77l(a)(2).
  11. Securities Act Rule 176, 17 C.F.R. § 230.176.
  12. Securities and Exchange Commission. In the Matter of Delphia (USA) Inc., Release No. IA-6539 (March 18, 2024).
  13. Securities and Exchange Commission. In the Matter of Global Predictions Inc., Release No. IA-6540 (March 18, 2024).
  14. Securities and Exchange Commission. Management’s Discussion and Analysis, Selected Financial Data, and Supplementary Financial Information, Release No. 33-10890 (November 19, 2020).
  15. Securities and Exchange Commission. Modernization of Regulation S-K Items 101, 103, and 105, Release No. 33-10825 (August 26, 2020).

Interested in analysis about the intersection of tech, policy and the law? Check out my Substack channel. https://theinnovationattorney.substack.com/