ChatGPT Legal: Is Using ChatGPT Legal for Lawyers in 2026?

Is ChatGPT legal for lawyers? Yes. ChatGPT is legal for lawyers in every US jurisdiction that has ruled on the question, on conditions that three authorities impose independently of one another: the vendor's usage policy, the ethics rules of the lawyer's own jurisdiction, and the standing order of the judge assigned to the matter. Each authority can permit what another restricts, so satisfying one of the three is not satisfying the other two.

This guide covers what the vendor policy says, what ABA Formal Opinion 512 requires, which bar authorities have ruled, what courts demand before an AI-assisted filing, and what a lawyer may bill when the tool does the work faster. The conditions apply to lawyers in law firms and to in-house legal teams alike.

[cta-1]

Key takeaways

Can lawyers use ChatGPT legally? Yes, subject to conditions set by three independent authorities.

1. The three authorities are the vendor's usage policy, the jurisdiction's ethics rules, and the assigned court's standing order. Satisfying one is not satisfying the others.
2. OpenAI did not ban ChatGPT from giving legal advice. The usage-policy clause restricts unsupervised advice from an unlicensed source, and it predates the consolidation that drew the coverage.
3. ABA Formal Opinion 512 creates no new rule. It reads the existing Model Rules onto generative AI use, and the supervisory duties are the ones that reach firm-level process.
4. On an hourly engagement, a lawyer bills the time actually expended, not the time the tool saved. That constraint reaches the fee arrangement before it reaches the timesheet.
5. Disclosure is court-specific. No general duty requires a lawyer to tell a court that AI assisted a filing, and a growing number of individual judges require exactly that.

Did OpenAI ban ChatGPT from giving legal advice?

No. OpenAI did not ban ChatGPT from giving legal advice, and the clause practitioners have been reading about does not restrict a lawyer who supervises the output. The usage policy prohibits the "provision of tailored advice that requires a license, such as legal or medical advice, without appropriate involvement by a licensed professional." The object of that restriction is unlicensed, unsupervised advice. A lawyer who reviews the output and takes professional responsibility for it is the involvement the clause requires.

The restriction is also not new. A publicly archived copy of OpenAI's usage policies from April 2024 already prohibited "Providing tailored legal, medical/health, or financial advice without review by a qualified professional and disclosure of the use of AI assistance and its potential limitations." The later consolidation restated long-standing language in shorter form, as Legal IT Insider's report on the usage-policy change describes. No new ban was introduced, and nothing a lawyer was permitted to do before the consolidation became impermissible after it.

What the clause actually reaches

The clause reaches unsupervised deployment: a firm that puts a ChatGPT-backed intake assistant in front of the public, answering questions about people's legal positions with no lawyer reading the answers. That is a materially different activity from a lawyer drafting a first pass and then editing it.

A usage policy and a model's behavior are different objects: the policy states what OpenAI permits, not what the product will produce in a given session, so a lawyer who reads it as a technical guardrail has misread it.

What ABA Formal Opinion 512 requires of a lawyer using ChatGPT

ABA Formal Opinion 512 requires a lawyer using ChatGPT to treat the tool as delegated assistance. The duties that already attach to delegated work apply unchanged. Issued by the American Bar Association's Standing Committee on Ethics and Professional Responsibility on July 29, 2024, it is the ABA's central guidance on a lawyer's use of generative AI, and it creates no new rule: it reads the existing Model Rules of Professional Conduct onto generative AI use and names the duty each one attaches.

The Model Rules the opinion reads onto generative AI use are:

1. Rule 1.1, competence
2. Rule 1.4, communication with the client
3. Rule 1.5, fees
4. Rule 1.6, confidentiality of information
5. Rule 3.3, candor toward the tribunal
6. Rules 5.1 and 5.3, the supervisory duties of partners and of lawyers directing non-lawyer assistance

Rule 1.4 answers whether the client has to be told, and the answer turns on the engagement rather than on the tool. Rule 1.6 governs what a lawyer may put into the tool in the first place, which the confidentiality section below takes up.

The supervisory duties under Rules 5.1 and 5.3

Rule 5.3 reaches a lawyer's responsibilities regarding non-lawyer assistance, and the opinion's reading makes a generative AI tool exactly that. A firm policy stating that AI use is permitted does not discharge the duty. The policy has to name who may use which tool on which matter types, because that is the level at which a supervising partner can be said to have made reasonable efforts to ensure the assistance is compatible with the lawyer's own obligations. A managing partner can act on a policy written at that level.

The opinion is not uncontested, and the contested part is the fee reasoning. The Thomson Reuters Institute's analysis of the fee guidance argues that the reasonableness requirement the opinion applies to flat fees creates more uncertainty than it resolves, which is why fees get their own section below.

Which bar authorities have ruled on lawyers using ChatGPT

Bar authorities that have ruled on lawyers using ChatGPT reach the same destination by different routes: generative AI is permitted, and the existing duties of competence, confidentiality, supervision, candor and reasonable fees apply to it unchanged. As AI use across the profession has grown, so has the number of authorities on record.

One correction to a claim that circulates widely, including in an earlier version of this guide: Michigan's generative AI ethics opinion, Judicial Ethics Opinion JI-155 of October 27, 2023, addresses judicial officers rather than lawyers, so it does not belong in a table of lawyer-facing authorities. The DC Bar opinion above is the lawyer-facing substitute. Two rows also sit outside the bar system and are included because they reach the same question: the New Jersey guidance came from the Judiciary rather than from a bar ethics committee, and the EU AI Act is a regulatory regime rather than an ethics authority.

The state bar opinions

California's guidance is among the most operationally detailed of the set, and it addresses billing directly. The State Bar's practical guidance on generative AI tells lawyers to understand the tool's limitations before relying on it, protect confidential information, review output for accuracy, and not charge clients for time the lawyer did not expend.

Florida takes the duty-mapping route instead. Ethics Opinion 24-1 permits generative AI provided the duties of competence, confidentiality, supervision of the tool as a form of assistance, and reasonable fees are met, and it is explicit that routing work to a tool does not reduce what the lawyer owes on any of those four duties.

The New York City Bar reaches generative AI through existing obligations rather than new ones. Formal Opinion 2024-5 applies competence, confidentiality, supervision, candor toward the tribunal, reasonable fees and client communication to generative AI without carving out anything technology-specific, a pattern later opinions follow.

The District of Columbia is the authority to read where a firm wants a lawyer-facing opinion with supervisory detail. Ethics Opinion 388 of April 2024 addresses competence, confidentiality, supervision and the accuracy of work product produced with generative AI, and it is the row that replaces the Michigan entry above.

Texas took a slower route. Rather than issue a single opinion, the State Bar convened a standing body to produce guidance, and the work of the Taskforce for Responsible AI in the Law treats understanding the tool, verifying AI-generated work, protecting confidential information and following court rules as the conditions of use.

Outside the United States

For a firm advising on matters that touch the European Union, the classification question comes before the ethics question. The EU AI Act attaches documentation, human oversight and conformity assessment obligations to systems inside the high-risk classification in Article 6, which turns on a system's purpose rather than on the profession using it, and a general-purpose assistant used for drafting is not the paradigm case.

What courts require before an AI-assisted filing

Court orders requiring disclosure of AI use in filings come from individual judges rather than from districts or from the federal rules. The duty is per-courtroom, so checking the assigned judge's standing order belongs in the filing workflow.

The scale of the problem driving those orders is measurable. Damien Charlotin's AI Hallucination Cases database recorded 2,125 decisions worldwide involving AI-generated content in court filings, 1,459 of them in the United States, when it was read for this guide on October 2, 2026.

What the standing orders actually require

The orders in circulation show the range of what individual judges have imposed:

1. Judge Brantley Starr, Northern District of Texas: certification that generative AI was not used, or that any AI-drafted content was checked by a human
2. Judge Michael Baylson, Eastern District of Pennsylvania: disclosure when AI helped prepare a filing, with confirmation that citations to law and to the record are accurate
3. Judge Gabriel Fuentes, Northern District of Illinois: disclosure of the tool and an explanation of how it was used
4. Judge Nina Wang, District of Colorado: an AI certification with every filing
5. Central District of California, several civil standing orders: a separate declaration when AI generated any part of a filing

The wording varies and the obligation does not. Most orders turn on human verification: confirming cited authorities exist, comparing quotations against the source material, and satisfying the certification Rule 11 already imposes. The sanctions narrative itself sits on the page about the lawyer fined over fake AI citations.

[cta-2]

How can lawyers use ChatGPT legally?

Lawyers can use ChatGPT legally by discharging four duties the ethics opinions and the standing orders impose between them: verify what the model cites, keep client material out of the account, supervise who uses which tool on which matters, and check the assigned court's order before filing. Each duty has a practitioner-level version more specific than the rule's own language.

Verify against the reporter, not the model

Verification means opening the authority, not asking the model whether the authority exists. A fabricated citation is usually internally consistent, so a follow-up prompt tends to confirm it. Purpose-built AI legal research tools narrow the problem without removing the step, which has three parts:

1. Pull the case from the reporter or the docket, not from the model's summary
2. Confirm the parties and the holding, not just that a case with that name exists
3. Compare any quoted passage against the source text word for word

Keep client material out of the account

The duty is discharged at input rather than at output. Redacting a name while pasting the rest of a disputed agreement leaves enough context to identify the parties, so the workable version is to keep the material out of the consumer account and move the work to a tool the firm has diligenced. The doctrinal question behind that choice, whether a conversation with a general assistant is protected at all, sits on the page about privilege and ChatGPT.

Classify the matter before the associate has to

Supervision under Rules 5.1 and 5.3 is a classification problem rather than a training problem, and the classification has to be made centrally. Where a firm has not decided which matter types a given tool may touch, the decision falls to whoever opens the document, which in practice means an associate weighing a confidentiality question against a filing deadline. The reviewable version of the duty records the call before the matter arrives and names the person who signs off on output leaving the firm.

Check the assigned court's order before filing

Because the duty is per-judge, the check belongs in the filing checklist next to local-rule formatting. Where a certification is required, the lawyer drafts it as part of the filing rather than adding it afterward, and where no order exists, the verification work still has to happen, because Rule 11 does not depend on a standing order.

ChatGPT legal client confidentiality concerns

Client confidentiality for lawyers using ChatGPT turns on one structural fact about the tool: a consumer ChatGPT account is a third party, and Model Rule 1.6 does not permit a lawyer to reveal information relating to the representation to a third party without the client's informed consent. Informed consent means consent given in advance, after the client has been told what the tool is and what happens to the material.

So the choice is narrow. A lawyer using a consumer account on client work needs one of two things:

1. Informed consent obtained from the client in advance
2. Material that does not relate to the representation at all

What the vendor retains, which account types carry which commitments, and what happens to chat history are treated on the page about ChatGPT and privacy. Whether those conversations can later be used against the account holder is a discovery question rather than an ethics one.

Can a lawyer bill for time ChatGPT saved?

A lawyer cannot bill for time saved using AI on an hourly engagement. ABA Formal Opinion 512 is direct about it: a lawyer who has agreed to bill by the hour bills the time actually expended, so a lawyer who needs fifteen minutes with the tool to produce a draft that would otherwise have taken far longer bills the fifteen minutes plus review time. The saved time belongs to the client.

If the tool compresses the work and the fee is tied to time, the fee falls, which is why the opinion reads inside firms as an argument for revisiting the engagement letter rather than for concealing the tool. Alternative fee arrangements are not a clean escape either: the opinion applies a reasonableness requirement to flat fees as well, so a firm charging the same fixed fee for work the tool now completes much faster may not be safe, which the Thomson Reuters Institute analysis identifies as the least settled part of the opinion.

What the opinion rules out billing for

Two costs are not billable:

1. Time spent learning a generative AI tool the lawyer will then use across many matters, which is professional development rather than work on any one client's matter
2. Any surcharge above the tool's direct cost, absent a disclosed agreement with the client

Both costs reach an invoice without anyone deciding to put them there, which is why the fee question belongs in the billing guidelines rather than in the AI policy.

The ethical considerations for lawyers using ChatGPT

The ethical rules for lawyers using ChatGPT are the rules that already governed delegated work, and New Jersey's guidance states how they transfer. The New Jersey Supreme Court's Preliminary Guidelines on the use of artificial intelligence by New Jersey lawyers, published in January 2024, acknowledge the benefits of the technology while addressing its tendency to produce confident and false output.

Two principles from the guidelines carry beyond New Jersey:

1. A lawyer must uphold diligence, honesty, client advocacy and confidentiality when using AI, which locates the duty in the lawyer rather than in the tool
2. Lawyers are not required to disclose their use of AI, which is a statement about the ethics rules and not about any particular court's standing order

Those two statements can diverge in the same matter. Where they do, the order controls what goes in the filing.

Where Spellbook fits

Spellbook and ChatGPT are different categories of tool for lawyers, and for the duties above the category difference is what matters. ChatGPT is a general assistant; Spellbook is an AI suite for transactional legal work that runs inside Microsoft Word and Google Docs, and the outputs are grounded in a team's own playbooks and precedents alongside 20M+ real contracts rather than generated from a general model's recollection. Grounding changes what verification costs, because a proposed redline that traces to a named standard is faster to check than one a lawyer has to confirm from scratch, which is the practical difference between the two routes to AI document review.

What a lawyer can diligence

On the confidentiality duty, the question is whether a lawyer can produce the documentation needed to diligence a tool before approving it for client work, which is the part a consumer account cannot supply. The commitments are published on our security page:

1. SOC 2 Type II certification
2. General Data Protection Regulation (GDPR) compliance
3. Zero data retention with the underlying model providers
4. Single sign-on (SSO) and multi-factor authentication (MFA) for account access

The Compare to Market feature measures a draft against 2,300+ benchmark standards, which turns a general sense of market terms into something a lawyer can cite internally.

Where it is not the right tool

The limitation is a category one: this is a commercial-contract specialist, so it is not the tool for the legal research or the litigation work a lawyer might otherwise take to a general assistant, and research-heavy teams are the ones most likely to need a second tool alongside it. The wider landscape, including the enterprise routes the large legal publishers now offer, is covered on the page about legal AI tools.

Final thoughts

Should lawyers use ChatGPT? The answer is narrower than whether they may, and it depends on which of the three authorities binds the matter in front of them:

1. The vendor's usage policy permits supervised professional use, and has done so since well before the consolidation that made the news
2. The ethics rules permit the use and attach the duties of competence, communication, confidentiality, supervision, candor and reasonable fees to it
3. The assigned judge's standing order is the only one of the three that can impose a disclosure obligation, and the one most often missed, because it changes from courtroom to courtroom

Lawyers who want the practical workflows rather than the rules will find them on the page about using ChatGPT in legal work. Try Spellbook free to see what grounded drafting looks like on a live agreement.

‍

Frequently asked questions

Do lawyers have to disclose when they use ChatGPT?

Lawyers do not have to disclose when they use ChatGPT as a general matter. Neither the ABA Model Rules nor the state bar opinions impose a standing duty to tell a court or a client that generative AI assisted the work. Disclosure becomes mandatory through three narrower routes:

1. A court's standing order
2. An agreement with the client
3. A jurisdiction-specific rule

A growing number of federal judges require disclosure, certification of human review, or both, so the obligation is decided by the assigned courtroom rather than by the profession.

Can ChatGPT give legal advice?

No, ChatGPT cannot provide legal advice. It lacks the licensed attorney's professional judgment, ethical responsibility, and contextual understanding required for proper legal counsel.

True legal advice requires professional judgment, understanding of jurisdiction-specific nuances, ethical responsibility to clients, accountability, and ongoing education about changing legal standards, none of which ChatGPT possesses.

Is a ChatGPT-drafted contract enforceable?

A ChatGPT-drafted contract is as enforceable as any other contract, because enforceability turns on the agreement rather than on the drafting method. The familiar elements still have to be present:

1. Offer
2. Acceptance
3. Consideration
4. Capacity
5. Legality

What the drafting method changes is the risk carried by the document's content, since an unreviewed draft can carry terms that do not say what the parties intended, omit a provision the deal needs, or misstate a governing law reference. That risk is the reason the lawyer reviews the draft.

Does a law firm need a written AI policy?

No rule requires a law firm to adopt a written AI policy, and the supervisory duties in Model Rules 5.1 and 5.3 are difficult to discharge without one. Several bar authorities, including the State Bar of California and the New Jersey Judiciary, encourage firms to adopt and periodically update internal policies. A policy that does the work names five things:

1. The approved tools
2. The matter types each tool may touch
3. Who reviews output before it leaves the firm
4. How client material is kept out of consumer accounts
5. When the policy is reviewed, since the bar guidance asks for periodic updating

Can lawyers ensure confidentiality when using ChatGPT?

Lawyers can protect confidentiality when using ChatGPT by not sharing sensitive client information with the chatbot. Or, they can ask for written client consent when necessary. If this is not feasible, lawyers should opt for reliable AI tools with strong security measures and zero data retention protocols.

Are there any clear legal regulations governing lawyers' use of ChatGPT?

Yes, there are clear legal regulations governing lawyers' use of ChatGPT in several states. The state bars of California, Florida, New York and the District of Columbia have published guidance that emphasizes the responsibilities of lawyers using AI: maintaining professional ethics and client confidentiality while overseeing AI outputs. As long as they comply with those rules, lawyers can use ChatGPT legally without a general duty to disclose it, unless the assigned court's standing order requires disclosure.

[cta-3]