The Artificial Intelligence Deceptive Advertising Act targets advertisements that use AI-generated or AI-modified synthetic content to create a false consumer impression. It is narrower than a general requirement to label all AI content.

Pennsylvania has enacted a new AI-specific advertising law. Senate Bill 806 was approved as Act 99 of 2026 on October 22, 2026, according to the General Assembly’s bill record. The law takes effect on December 21, 2026—60 days after enactment.

The controlling final text is Printer’s Number 1808, which names the statute the Artificial Intelligence Deceptive Advertising Act.

What the law prohibits

The statute does not require a disclosure every time AI appears in advertising. Its trigger is more specific.

An advertiser may not knowingly, recklessly, or willfully create and disseminate, or cause to be created and disseminated, an advertisement using synthetic content in a way that creates a false consumer perception about the nature, origin, or authenticity of the goods or services being advertised, unless the advertisement carries a clear and conspicuous disclosure.

The required disclosure must state that the advertisement contains content modified or generated using artificial intelligence. Under the statute, it must appear when the synthetic content is first presented to the consumer, be delivered in the same medium as that content, and—where technically feasible—be permanent or extraordinarily difficult to remove.

The law defines synthetic content as information, including an image, video, audio clip, or text, that has been significantly modified or generated by AI. It also supplies detailed standards for when a disclosure is “clear and conspicuous,” including requirements tied to visibility, audibility, proximity, language, and legibility.

What the law leaves outside its disclosure rule

Act 99 expressly excludes three categories from the disclosure provision:

• content that is evidently fictional from the content or context;
• parody, satire, commentary, or criticism; and
• content limited to text.

Those exclusions, together with the false-consumer-perception requirement, matter. The statute is not a blanket AI-labeling law, and it does not say that every AI-generated image, video, audio clip, or other synthetic element in an ad must be labeled.

The final text also does not define “advertiser.” How that term applies in particular arrangements may therefore require further interpretation as the law is enforced.

Enforcement and remedies

The Attorney General or an appropriate district attorney may seek a temporary or permanent injunction when there is reason to believe a person is using, or is about to use, a prohibited practice and proceedings would be in the public interest. Those officials may also bring a civil action for violations.

A court of competent jurisdiction may award a civil fine of up to $3,000, along with actual attorney fees and court costs. The statute does not specify whether the $3,000 maximum applies per advertisement, per violation, or on another basis. It also does not expressly create a private right of action.

The law further states that disclosure under the Act does not limit other recourse available under law.

The practical compliance question

For businesses, agencies, and advertisers using AI-altered or AI-generated material, the question is not simply whether AI was used. The statutory question is whether synthetic content is used in an advertisement in a manner that creates a false consumer perception about the advertised goods’ or services’ nature, origin, or authenticity.

Where the rule applies, the disclosure must be designed for the medium in which consumers encounter the synthetic content. A visual disclosure that is too brief to read, an audio disclosure that cannot be understood, or a disclosure separated from the altered content may not satisfy the statute’s stated standards.

Act 99 gives Pennsylvania a targeted consumer-advertising rule for deceptive uses of AI-generated or AI-modified content. Its reach will become clearer through enforcement and any future guidance, but its effective date is already set: December 21, 2026.

Sources

• palegis.us
• palegis.us

Wikimedia Asks the OpenAI Copyright Court to Separate AI Training From Retrieval

The filing proposes different fair-use analysis for pre-training and retrieval-augmented generation. It is an amicus request, not a court ruling; ECF No. 2361 does not itself resolve whether leave will be granted.

The Wikimedia Foundation has asked for permission to file an amicus brief in In re: OpenAI, Inc. Copyright Infringement Litigation, the Southern District of New York copyright MDL that includes The New York Times litigation against OpenAI and Microsoft.

Wikimedia's October 29 filing, ECF No. 2361, asks to submit a proposed brief in support of neither party. The immediate procedural point matters: the filing itself does not grant leave or adopt Wikimedia's proposed fair-use analysis. It is a request to be heard, not a decision about AI training, retrieval, or copyright liability.

Still, the brief offers a useful map of a distinction likely to matter across AI copyright disputes: whether a system is using copied material during model development or retrieving and delivering material to a user at runtime.

The proposed distinction is use-by-use

Wikimedia argues that fair use should be assessed separately at different stages of an LLM-based chatbot's development and operation. Its proposed brief describes that as a use-by-use inquiry rather than a single answer for every copy made in building or operating an AI product.

On Wikimedia's account, the pre-training use of copied works is likely fair use when it helps create a model that is different from the source works. That is an argument in the proposed brief, not a holding. It does not establish that a court will treat any particular training dataset, acquisition method, or model output as fair use.

The proposed brief takes a different position on retrieval-augmented generation, or RAG. It describes RAG and similar grounding techniques as systems that can retrieve and copy material from the web to answer a user's request. Wikimedia argues that this use should receive its own fair-use analysis and is likely not fair under the four-factor test when it displaces access to the source material.

That distinction does not turn on whether the source is available online. It is about the particular use at issue. The filing argues that a court should not treat every technical step involved in an AI system as legally identical merely because the same works may be involved.

Why Wikimedia says the split matters

Wikimedia's request connects the proposed analysis to the economics and governance of open knowledge. The organization says its projects depend on Creative Commons licensing terms, including attribution and ShareAlike conditions, and on human contributors continuing to create and improve material.

The motion says Wikimedia has a distinct perspective because its services sit at the intersection of freely licensed knowledge, machine-learning tools, and the traffic demands created by automated scraping and crawling. Its proposed brief argues that an undifferentiated fair-use rule could miss how retrieval and similar grounding systems may direct users away from a source while reproducing or serving its material.

Those are Wikimedia's positions. They do not establish that an AI system has infringed copyright, that every RAG implementation uses content unlawfully, or that all model training is protected by fair use. The MDL court will decide the questions properly before it on the record and arguments it accepts.

The filing is narrower than a ruling—and broader than a technical footnote

The procedural posture is narrow. Wikimedia asked Judge Sidney H. Stein for leave under the court's amicus process and attached a proposed 5,831-word brief. The motion states that no party or party counsel authored the proposed brief and that no person other than Wikimedia funded its preparation or submission.

But the distinction it asks the court to consider reaches beyond a single feature label. AI providers increasingly combine pre-trained models with search, retrieval, citation, browsing, or grounding layers. A fair-use analysis that separates those functions could make system design and deployment details more legally significant than a general description of a model as simply “trained on” online content.

For publishers and other rights holders, the proposed split focuses attention on whether a service is substituting for access to the underlying work. For AI companies and open-knowledge projects, it raises a parallel question about how copyright analysis should account for model development, current retrieval, attribution, licensing terms, and the practical burden that automated access places on a source.

None of those questions is resolved by ECF No. 2361. A later order on the motion would change the posture; a merits ruling would be a separate development altogether.

Filed motion and proposed brief: ECF No. 2361 Public docket snapshot: In re: OpenAI, Inc. Copyright Infringement Litigation Wikimedia's announcement: Wikimedia Foundation

Montana’s AI Election-Deepfake Law Remains Enacted. A Federal Court Has Blocked Enforcement Against Two Challengers.

Montana’s election-deepfake statute is still on the books. But a federal court has temporarily barred state officials from enforcing it against the two plaintiffs in a First Amendment challenge—and expressly declined to issue statewide relief.

Montana’s SB 25, enacted as Chapter 541 in 2025, is an election-deepfake disclosure law. The statute does not simply announce a general prohibition on “deepfakes.” It creates a defined disclosure-and-enforcement regime for specified election and electioneering communications involving a covered deepfake of an on-ballot candidate or political party within 60 days of the initiation of voting in the qualifying election.

That regime now sits alongside an important litigation limit. On September 16, Judge Susan P. Watters of the U.S. District Court for the District of Montana granted a preliminary injunction in Accountability in State Government v. Knudsen, No. 6:26-cv-00038-SPW. The court found that the plaintiffs, Accountability in State Government and Dan Bartel, had shown a likelihood of success on their First Amendment challenge and met the other requirements for preliminary relief.

The central qualification is the one most likely to get lost in a headline: the injunction protects the named plaintiffs. It does not invalidate SB 25 statewide or prevent Montana from enforcing it against everyone else.

What SB 25 regulates

The statute defines “AI-generated content” as content substantially created or modified by generative artificial intelligence in a way that materially alters its meaning or significance to a reasonable person; minimal edits that do not materially alter meaning or significance are excluded. Its definition of “deepfake” centers on AI-generated or synthetic media depicting a candidate or political party, made with intent to injure reputation or deceive a voter, that appears to show speech or conduct that did not occur or conveys a fundamentally different impression to a reasonable person.

For a covered election or electioneering communication within the statutory 60-day period, SB 25 generally requires a disclosure if a person working in an official election capacity pays for or sponsors the production, creation, or distribution of a covered deepfake that the person knew or reasonably should have known was covered. The required disclosure states:

“This _____ (image/audio/video/multimedia) has been significantly edited by artificial intelligence and depicts speech or conduct that falsely appears to be authentic or truthful.”

The law specifies how that disclosure must appear in different media. For example, it sets particular rules for print, video, internet graphics or text, and audio. It also includes stated exceptions, including for certain bona fide news reporting, broadcasters in specified circumstances, satire or parody, platforms that demonstrate their disclaimer requirements satisfy the statute’s disclosure conditions and that they supplied those requirements to the purchaser, and interactive computer or telecommunications services carrying third-party content.

Those details matter. A law described casually as an election-deepfake “ban” may operate very differently from a disclosure statute with definitions, timing triggers, medium-specific requirements, and exceptions. Any coverage or compliance analysis has to begin with the actual statutory elements rather than the label alone.

The preliminary injunction is deliberately plaintiff-specific

The court’s filed order grants preliminary relief, not final judgment. At this stage, Judge Watters concluded that the plaintiffs had demonstrated a likelihood of success on their First Amendment claim and had satisfied the other preliminary-injunction factors.

But the court rejected the plaintiffs’ request for a universal injunction. It reasoned that the requested relief had to be evaluated under the Supreme Court’s then-controlling limits on injunctions benefiting nonparties, discussed in the order through Trump v. CASA, Inc. The court stated that it lacked authority to issue a universal injunction prohibiting enforcement “globally against anyone impacted,” and held that the relief “must be limited to the named Plaintiffs.”

The operative language is correspondingly narrow. The order enjoins Attorney General Austin Knudsen, Lewis and Clark County Attorney Kevin Downs, and Commissioner of Political Practices Chris Gallus from reviewing, investigating, prosecuting, adjudicating, or enforcing the Act against the named plaintiffs.

That distinction is substantive, not technical. The ruling does not establish that the statute is finally unconstitutional on its face, and it does not create a general safe harbor for campaigns, committees, platforms, or other speakers who are not parties to the case. The statute remains enacted; the order leaves the nonparty enforcement posture to later proceedings and facts not resolved by this injunction.

Why the scope matters for election-speech planning

The decision illustrates a recurring problem in AI-election-law coverage: an early constitutional ruling can have real practical importance without changing the rule for every speaker.

For the named plaintiffs, the preliminary injunction bars the identified officials from using the statute against them while the case proceeds. For others, analysis still begins with SB 25’s text: whether a planned communication falls within the statute’s definitions, timing window, disclosure rule, or an enumerated exception. The order does not extend its protection to nonparties or resolve whether their communications satisfy the statute’s coverage requirements or exceptions.

That does not mean the injunction is unimportant. The court’s likelihood-of-success finding flags constitutional questions that may shape the merits case, future enforcement decisions, and challenges to similar laws. It also shows why the remedy question can be as consequential as the merits question: even where a court sees a serious First Amendment issue, the scope of preliminary relief may turn on who is before the court and what relief the court believes it can lawfully award.

The litigation remains active

The public docket shows that the case has continued beyond the September 16 order. It lists a preliminary pretrial conference order on September 29, a joint case-management plan on October 27, and a transcript-order form on October 28. A docket check on October 29 identified no later merits decision or appeal entry.

The next meaningful development could be a merits ruling, a change in the injunction’s status, an appeal-related filing, or a record that clarifies how the statutory regime is being applied outside this dispute. Until then, the durable takeaway is precise: Montana’s SB 25 is still enacted, and the federal court’s preliminary order protects these plaintiffs—not the public at large.

Sources

• Montana SB 25 / Chapter 541 (enrolled act)
• Accountability in State Government v. Knudsen, preliminary-injunction order, Doc. 31 (D. Mont. Sept. 16, 2026)
• Public docket mirror for Accountability in State Government v. Knudsen, No. 6:26-cv-00038-SPW (D. Mont.)

The UK’s Over-£700 Million AI Budget Package Is a Security Bet, Not a New Compliance Regime

Britain’s 28 October Budget puts substantial money behind AI security, resilience and domestic capability. It does not, by itself, create new AI rules, private-sector duties or compliance deadlines.

The UK government’s Budget 2026, published on 28 October, commits over £700 million across AI measures. Around £500 million is directed at the “secure foundations” for an AI-enabled economy: security, resilience and sovereign capability.

The package is consequential as industrial and security policy. It is not, however, a new AI compliance regime.

The Budget does not set out an operative AI regulation, a private-sector duty, an effective date or a compliance deadline. Organisations should not mistake a large public investment package for a rulebook.

A shift toward security capability

The Budget presents AI as both an economic opportunity and a source of cyber, biosecurity and other risks. Its answer is institutional capacity and public investment.

Among the measures, the government says it will provide:

• an additional £120 million for the AI Security Institute (AISI);
• £70 million for a National Exposure and Real-Time Vulnerability Centre at the National Cyber Security Centre;
• £60 million for the first phase of a National Cyber Shield;
• £40 million for a National Agentic Security Unit;
• £100 million to establish an AI Biosecurity Lab, alongside further funding for bio-surveillance and medical-countermeasure development; and
• additional funding to harden public-sector technology and support AI-related cyber defence.

Taken together, the measures show a government placing AI security closer to national resilience, cyber defence and strategic industrial capability. The Budget also links the package to wider efforts to build domestic AI capacity, including support for UK firms, chip design and strategic data-centre policy.

That framing matters. It suggests that UK AI policy is being developed not only through discussions of governance and safety, but through the institutions expected to test, secure and respond to emerging technological risks.

The AISI is being funded, not formally remade

The Budget describes the AISI as the AI Security Institute and provides additional funding to strengthen its work on understanding AI capabilities and advancing technical practices for secure AI.

That is significant support for the institution. It is not the same thing as a published instrument formally expanding AISI’s remit, powers or legal role.

The distinction is worth preserving. A budget commitment can signal political priority and enable new work, but it does not itself define the legal boundaries of an institution’s authority. Any claim of a formal remit change would need to rest on a separate official instrument.

Cyber Shield remains a proposal in development

The National Cyber Shield is one of the package’s more closely watched elements. The Budget allocates £60 million to its first phase and says it will develop agentic capabilities for cyber defence with the UK cyber-security industry.

It also refers to “scoping for a regulatory deployment sandbox” to help organisations deploy new technologies.

That is not evidence that a sandbox is already operating, that its rules have been published, or that autonomous agents are already deployed into critical national infrastructure. The Budget describes work in development, not a completed regulatory framework.

The same restraint applies to the new National Agentic Security Unit and the NERV Centre. The Budget identifies funding and intended capabilities, but it does not publish operating rules, start dates, statutory powers or implementation documents for those programmes.

From pre-announcement to Budget posture

An HM Treasury announcement of 24 October previewed the AI package before the Budget. It described planned investments in areas including the AI Security Institute, cyber defence and biosecurity.

The 28 October Budget is the more authoritative source for the package’s final posture. It confirms the broader investment total and gives the clearest official account of the security-and-resilience measures.

The 24 October announcement did not state a final package total. The Budget supplies it: over £700 million across AI measures, with around £500 million for the secure foundations of an AI-enabled economy.

What the package means for businesses

The immediate implication for businesses is strategic rather than compliance-led.

Companies working in AI, cyber security, infrastructure, health technology and related sectors may find new government programmes, procurement opportunities, partnerships or technical initiatives emerging from the Budget’s commitments. The details will depend on programme documents and implementation decisions that have not yet been published.

What businesses should not infer is a new general obligation to adopt specific AI controls, participate in a sandbox, report agentic incidents or meet a new AI-security deadline. Those propositions are not established by the Budget.

The policy direction is clear: the UK wants stronger domestic capability to test AI systems, secure public and critical infrastructure, and respond to AI-enabled threats. The legal consequences are not yet equally clear, because the Budget does not create them.

The next meaningful signals will come from the implementation record: published sandbox rules, programme guidance, operational documents and any later legislation or regulation. Until then, this is a substantial public-policy commitment—not a compliance alert.

Sources

• HM Treasury, Budget 2026 (28 October 2026)
• HM Treasury, Chancellor unveils Britain’s ambitious plan for AI (24 October 2026)

Trump’s “Super Intelligence” Executive Order Changes Terminology in Specified Non-Statutory Executive-Branch Materials

Executive Order 14434 directs executive agencies to use “Super Intelligence” and “SI” in specified non-statutory materials. It does not create a new technology category or a general private-sector compliance obligation.

President Donald Trump signed Executive Order 14434, “Inaugurating the Era of Super Intelligence”, on September 29, 2026. Published in the Federal Register on October 2, the order begins a terminology shift across parts of the executive branch—but it does not legally rename artificial intelligence throughout federal law.

The distinction is easy to lose in the rhetoric surrounding the order. In the covered executive-branch materials, “Super Intelligence” and “SI” are the required substitutes to the maximum extent permitted by law. Section 3(a) uses the existing statutory definition of artificial intelligence for the order’s purposes; the order does not amend the statute.

For organizations that work with federal agencies, the immediate consequence is largely practical: expect to see “SI” in covered new agency materials. The order itself does not, however, impose a direct duty on private companies, federal contractors, or grant recipients to adopt the term.

What the order directs

Section 2(a) instructs executive departments and agencies, “to the maximum extent permitted by law,” to use “Super Intelligence” and “SI” in place of “Artificial Intelligence” and “AI” in official correspondence, public communications, websites, reports, policy documents, and other non-statutory documents within the executive branch.

That is a meaningful directive inside the executive branch. Agencies have already used the terminology in official materials, including materials associated with NASA and the National Science Foundation.

But the order’s scope is narrower than a universal renaming. Section 2(b) says that nothing in the section requires alteration of previously issued regulations, presidential actions, contracts, grants, or other historical documents. Nor does the order amend statutes enacted by Congress.

The controlling text is available in the Federal Register and in the official GovInfo PDF.

“SI” still means AI for the order’s purposes

Executive Order 14434 does not establish a technically or legally distinct category of technology called “Super Intelligence.”

Instead, Section 3(a) says that, for purposes of the order and except where otherwise provided by law, “Super Intelligence” and “SI” mean the technologies and systems encompassed by “artificial intelligence” as defined in 15 U.S.C. § 9401(3). In other words, the terminology changes, but the operative definition used by the order does not.

That matters for legal analysis. A company cannot assume that a federal statute, regulation, contract, grant condition, or procurement rule using the term “artificial intelligence” has been amended merely because an agency begins using “SI” in new materials.

The executive order is a presidential directive, not notice-and-comment rulemaking and not legislation.

No new general compliance deadline for private organizations

The order includes one date that may attract attention. Section 3(b) gives the Assistant to the President for Science and Technology 60 days to submit proposed legislative language to the President. Measured from the September 29 signing date, that internal deadline falls on November 28, 2026.

It is not a private-sector compliance deadline. It is also not an agency-wide deadline for completing terminology changes.

The contemplated submission could lead to proposed legislation, but it cannot itself amend the U.S. Code. Any statutory change would require congressional action through the ordinary legislative process.

The order does not set out a comprehensive transition schedule for agencies or define every “applicable setting.”

What federal-facing teams should do now

Government-facing policy, procurement, grant, and compliance teams should be prepared to encounter “SI” in new federal communications and executive-branch documents.

That is not the same as a requirement to change a company’s own terminology. A private obligation would need to appear in the relevant instrument—for example, a solicitation, contract clause, grant notice, award term, or agency-specific policy.

The safest near-term approach is straightforward:

• Treat “SI” as the executive order’s preferred terminology for the existing statutory AI concept.
• Review agency-specific documents for any actual contractual, grant, or regulatory obligations.
• Do not assume that the executive order retroactively changes prior agreements or legal texts.
• Do not treat the November 28 White House deadline as an external compliance date.

Executive Order 14434 may be the start of a broader policy and legislative effort. For now, its clearest legal effect is narrower: it directs a terminology change in specified executive-branch materials while leaving the existing statutory definition of artificial intelligence in place.

Norway government announces intent to propose temporary AI-glasses use ban at selected places

Norway’s government has said it plans to return to the Storting with a proposal restricting the use of AI-enabled glasses in certain locations. The proposal has not yet been introduced, and a separate parliamentary initiative would go further by covering sales and a broader class of wearable surveillance technology.

Norway has not banned AI glasses. But the government has moved beyond its earlier call for organizations to consider internal policies and now says it intends to return to the Storting with a proposal for a temporary, place-limited restriction on their use.

In an October 5 announcement, the Ministry said the government plans to return with a proposal for a temporary prohibition on using AI glasses in selected places. Minister Torgeir Micaelsen also confirmed that intended course during an October 5 Storting debate.

The announcement is a material step beyond the government’s August 25 announcement describing an expert group and encouraging organizations to develop internal policies. It signals an intention to return with a proposal. It does not itself create a new legal duty.

As of October 27, no government bill text, consultation text, committee recommendation, enacted law, effective date, or compliance deadline had been identified in the reviewed official materials. That is a statement about the official materials reviewed for this article, not a claim that no later document could exist elsewhere.

A proposed restriction on use, not a sales ban

The government’s stated direction is narrower than a total ban on AI glasses. It concerns their use in selected locations, rather than their sale or ownership.

The Ministry has identified possible settings that could include public-facing, child-centered, and privacy-sensitive places. Those examples should not be treated as settled legal coverage. Any eventual proposal would need to clarify the covered devices, locations, conduct, duration, exceptions, enforcement arrangements, and whether penalties would apply.

The government has also indicated that exceptions may be considered for vulnerable groups and for limited socially beneficial uses. The scope and conditions of those exceptions remain unknown.

That leaves organizations with a planning question rather than a new compliance obligation. Employers, schools, healthcare providers, retailers, and operators of public venues may have reason to watch the proposal closely, particularly where wearable cameras or sensors could create privacy and surveillance concerns. But they should not treat the October 5 announcement as an operative statutory regime.

Parliament is considering a broader opposition proposal

A separate proposal from Socialist Left Party representatives Kirsti Bergstø and Mirell Høyer-Berntsen would go substantially further.

Their Document 8:8 S (2026–2027) asks the Storting to direct the government to bring forward proposals for a temporary ban on both sales and use of body-worn technology equipped with cameras or other sensors that can enable covert surveillance.

That formulation is broader in two ways. It would cover sales as well as use, and its operative request is not limited to products marketed as AI glasses. The proposal instead reaches a wider category of body-worn devices capable of covert surveillance, while excluding assistive devices for people with disabilities.

The matter was referred on October 15 to the Standing Committee on Local Government and Public Administration. Its case page showed committee consideration as of October 27, without a debate or decision.

The proposal is therefore not law, and its filing does not independently impose obligations on businesses or consumers.

Separate sales-related motions have already failed

The parliamentary record also shows that two separate KrF motions concerning AI-glasses sales were rejected on October 6. The official voting transcript records votes of 81–19 and 82–19.

Those votes do not resolve the pending Document 8:8 S proposal or the government’s announced legislative plans. They do show that parliamentary approaches to AI-glasses regulation are taking different forms: specific sales restrictions, a broader sales-and-use proposal, and the government’s narrower stated plan for place-based limits on use.

What remains unsettled

The legal details that would determine the practical effect of any future restriction are still absent. In particular, official materials reviewed through October 27 did not establish:

• which devices would be covered;
• which locations or conduct would be restricted;
• the measure’s legal basis;
• the responsible enforcement authority or any penalties;
• the duration or commencement date of a temporary measure;
• the full set of exceptions; or
• the membership and formal terms of reference of the government’s expert group.

Those gaps matter. A restriction limited to specified uses in specified places would present a different legal and operational question from a prohibition on sale, possession, or all use of a broad category of wearable technology.

For now, Norway’s relevant developments are a government intention to return with a proposal and a pending parliamentary initiative—not an enacted AI-glasses ban, a new binding duty, or a compliance deadline. The government’s announced intention to return with a proposal is not itself an established legislative measure.

Florida appellate court imposes sanctions over unreliable AI-assisted briefing

The Florida Fourth District Court of Appeal has affirmed summary judgment and granted appellate sanctions after finding that an initial appellate brief contained fabricated cases, fictional quotations, inaccurate citations, and authorities that did not support the propositions for which they were cited.

In its October 7, 2026 opinion in Kid International, LLC v. City National Bank of Florida and 2655 SW 25, LLC, No. 4D2025-2599, the court granted sanctions under Florida Rule of Appellate Procedure 9.410(b).

The opinion states that the brief cited Ogilvie v. Sebring Airport Authority seven times, including a quotation attributed to that case. The cited case was fabricated or nonexistent. The court also identified inaccurate citations and descriptions of holdings, fictional quotations from actual cases, and authorities that did not support the cited arguments.

Counsel told the court that generative AI had been one component of the research and drafting process. The court’s opinion focused on the resulting failure to verify the legal authorities and quotations presented in the filing.

What the opinion establishes

The Fourth District:

• affirmed the summary judgment under review;
• granted sanctions under Rule 9.410(b);
• awarded reasonable appellate attorney’s fees against the appellants and their counsel, jointly and severally; and
• remanded for determination of the amount of those fees.

The opinion does not set a dollar amount for the fee award; it remanded for the amount to be determined.

The court also included a mandatory instruction concerning a potential rehearing motion: “If a motion for rehearing is filed in this court, then services rendered in connection with the filing of the motion, including, but not limited to, preparation of a responsive pleading, shall be considered in computing the amount of the fees.”

The opinion further states: “Not final until disposition of timely-filed motion for rehearing.” That notice makes the opinion’s finality conditional when issued; this article does not infer from the opinion alone whether a rehearing motion was later filed or resolved.

Why the ruling matters for AI-assisted legal filings

The decision is a pointed example of the verification risks associated with generative-AI-assisted legal research and drafting. The problem identified by the court was not simply that AI was used. Rather, counsel presented legal propositions, cases, and quotations that were not reliably checked before being filed.

For litigants and lawyers, the opinion underscores that a brief’s authorities must be accurate, real, and supportive of the argument for which they are offered. A filing can appear conventionally researched while still containing invented or distorted authority if AI-generated output is not independently verified against primary legal sources.

The ruling does not announce a general ban on AI use or establish a standalone new professional rule governing AI. It applies existing appellate-sanctions authority to the briefing conduct before the court.

The article is based on the published Fourth District opinion. The related Florida ACIS docket is linked in the metadata, but its docket entries were not text-extractable for this review.

Georgia’s New AI Guardrail for Health-Coverage Reviews Takes Effect January 1, 2027

Effective January 1, 2027, Act 411 will permit AI-supported utilization review, but an artificial intelligence system, artificial intelligence, or other software tool may not issue an adverse determination to a patient until a natural person qualifying as a private review agent or a utilization review entity conducts a utilization review in which a clinical peer participates.

Georgia’s SB 444 became Act 411 when it was signed on May 5, 2026. The Act takes effect January 1, 2027.

When effective, the Act will add O.C.G.A. § 33-46-7.1 to Georgia’s insurance code. It does not prohibit artificial intelligence in utilization review. Instead, it will permit private review agents and utilization review entities to use artificial intelligence systems, artificial intelligence, or other software tools, provided that the systems or tools are part of a utilization-review plan that complies with Chapter 33-46 and rules and regulations adopted by the Insurance Commissioner. The tools may be used to automate tasks, reduce administrative burdens, participate in decision-making processes, and perform other lawful functions, subject to the Act’s restrictions.

The line the statute draws concerns adverse determinations.

Under the enacted text, AI systems, AI, and other software tools:

“shall not issue an adverse determination to a patient until a natural person qualifying as a private review agent or a utilization review entity conducts a utilization review in which a clinical peer participates.”

The statute also states that those systems and tools may not supersede the clinical peer’s judgment.

Chapter 33-46 defines an “adverse determination” as “a determination based on medical necessity made by a private review agent or utilization review entity not to grant authorization to a hospital, surgical, or other facility or to a healthcare provider’s office for admission, extension of an inpatient stay, or a healthcare service or procedure.” In this article, the term refers only to that statutory category, not to every coverage denial or adverse healthcare decision.

What the law does—and does not—say

The final as-passed text does not establish a general human-oversight rule for every use of AI in healthcare administration. The new Code section is written in Chapter 33-46 terms and must be read with that chapter’s applicability provisions, definitions, and exclusions.

When effective, the new section will permit private review agents and utilization review entities to use artificial intelligence systems, artificial intelligence, and other software tools within their utilization-review plans, subject to Chapter 33-46 and the Commissioner’s rules.

Those systems and tools may not issue an adverse determination to a patient until a natural person qualifying as a private review agent or a utilization review entity conducts a utilization review in which a clinical peer participates. They also may not supersede that clinical peer’s judgment.

For purposes of the new Code section, “artificial intelligence” means a machine-based system that can make predictions, recommendations, or decisions influencing real or virtual environments for human-defined objectives. The section separately defines an “artificial intelligence system” as an engineered or machine-based system that can receive information and use it to emulate a human cognitive process, including learning, generalizing, reasoning, planning, predicting, acting, or communicating.

January 1 is the operative date

Act 411 does not state a separate compliance deadline. Its operative date is January 1, 2027.

That gives private review agents and utilization review entities time to identify where AI or other software tools in their utilization-review plans issue adverse determinations to patients and whether their workflows provide for the required natural-person utilization review and clinical-peer participation before issuance. Other organizations should analyze applicability under current Chapter 33-46 rather than assume that the new section covers every healthcare workflow.

The new section should not be read in isolation. Before treating Act 411 as a complete implementation checklist, affected organizations should consult the current applicability provisions and definitions in O.C.G.A. Chapter 33-46, the applicable Georgia Insurance Commissioner rules, and any relevant agency guidance.

The central takeaway is straightforward: effective January 1, 2027, Georgia will expressly permit private review agents and utilization review entities to use AI and other software tools in covered utilization review, but the tools may not issue an adverse determination to a patient before the required natural-person utilization review with clinical-peer participation, and they may not supersede the clinical peer’s judgment.

Z.ai’s GLM-Legal Puts a Legal-Specific Model Into the Market

Z.ai has launched a beta legal-specific model, GLM-Legal, built on its GLM-5.3-Flash foundation model. The company says it has added legal-domain mid-training and reinforcement-learning-based post-training aimed at work such as contract review, evidence analysis, legal research, drafting, and compliance planning.

The announcement is more than a general-purpose model vendor adding a legal prompt library. Z.ai is presenting GLM-Legal as a separately trained legal model and plans to bring it into AutoClaw, its desktop-native agent platform. It is also inviting law firms, legal-technology companies, in-house teams, and independent practitioners to apply for beta API access.

That positioning deserves attention. But the launch materials leave important questions unanswered for prospective legal users: where customer data will be processed and retained, what controls will govern access and model improvement, what the commercial terms will be, and how the system will perform on a firm’s own matters and jurisdictions.

What Z.ai Says It Built

According to Z.ai, GLM-Legal uses a two-stage training approach. The company says it first expanded the model’s exposure to legal corpora and legal concepts through “mid-training,” then used reinforcement learning across what it describes as tens of thousands of realistic legal environments to improve task execution.

The claimed goal is not simply to produce fluent legal text. Z.ai says the model is intended to ingest source records, connect facts to doctrine, structure an argument, and deliver work in a requested format. Its examples include marking up a software-services agreement, comparing inconsistent project records in a dispute, and building a cross-border privacy-compliance checklist.

Those examples should be read as product demonstrations, not proof that the model reaches correct conclusions in a live matter. In particular, contract positions, litigation theories, and country-specific compliance advice depend on the governing law, the underlying documents, and human legal judgment. The company’s examples do at least make its intended workflow clear: the system is meant to operate on a matter record and produce a concrete work product, rather than only answer a standalone question.

The Reported Benchmark Results

Z.ai reports that GLM-Legal achieved a 19.2% all-pass rate on the Harvey Legal Agent Benchmark, improving by 12.5 percentage points over its base model. It also says the model matched frontier closed-source systems on LegalBench and PLawBench. The company says Vals independently evaluated the Harvey benchmark and LegalBench results.

Those figures are useful signals, but they do not settle a purchasing decision. An all-pass metric is demanding precisely because partial success may not be enough for a real workflow. At the same time, a benchmark score does not establish how a tool will behave on a firm’s document types, privilege practices, governing law, terminology, or review controls.

The most useful follow-up from Z.ai would be a fuller evaluation record: benchmark versions and task composition, comparison models and settings, failure categories, citation-quality results, and repeatable methods. Legal teams should also ask for a controlled pilot using representative—but appropriately authorized—materials and a review rubric that measures factual support, issue spotting, citation verification, confidentiality handling, and the amount of lawyer rework required.

A Different Kind of Legal-AI Market Entry

The legal-AI market has often been divided between general-purpose model providers and legal-technology companies that build products on top of them. GLM-Legal sits closer to a third position: a model developer offering a legal-tuned model directly while planning to integrate it into its own agent platform.

That vertical integration could matter if the model, retrieval, tool use, document handling, and user interface are designed and evaluated together. Z.ai’s launch page emphasizes that its model training and “harness” evolve together, with production feedback intended to influence both the model and the task environment.

It also concentrates accountability questions. A legal buyer will want to know which entity supplies the model, hosts the data, runs the agent layer, and bears responsibility for security commitments, incident response, access controls, retention, and support. A legal-specific model does not remove those questions; it makes them central to the product evaluation.

The Questions Before Production Use

Before using GLM-Legal for client or sensitive internal work, a firm or legal department should obtain written answers on at least these points:

• Data location and access: Where are prompts, uploads, outputs, logs, and backups processed and stored? Which personnel, subprocessors, or affiliates can access them?
• Training and retention: Are customer materials used to train or improve any model? Can the customer opt out, and what are the deletion and retention controls?
• Security and identity: What encryption, authentication, role-based access, audit logging, and incident-notification commitments apply?
• Matter controls: Can the product segregate workspaces and clients, preserve source-document links, and support lawyer review before any work product is used externally?
• Jurisdictional performance: Which languages, legal systems, and practice areas have been evaluated, and which remain outside the supported scope?
• Evaluation and redress: Can the vendor provide reproducible evaluation material, explain known limitations, and commit to a process for reporting serious errors or security issues?

These are ordinary legal-technology diligence questions, not a finding that GLM-Legal lacks the relevant safeguards. The public launch material does not appear to answer them in detail. Until it does, a beta evaluation should be bounded, supervised, and separated from unreviewed client-facing work.

The Practical Takeaway

GLM-Legal is a notable signal that legal-specific model development is not confined to established Western legal-information and legal-tech vendors. Z.ai claims that legal-domain training and agent-oriented post-training enabled GLM-Legal to match frontier systems on selected benchmarks despite what the company describes as a significantly leaner parameter footprint.

For buyers, the headline should not be “legal-specific” alone. The meaningful test is whether the product can show reliable performance on authorized representative work while meeting the organization’s requirements for confidentiality, data governance, supervision, auditability, and jurisdictional fit. Z.ai’s beta launch creates an opportunity to test that proposition. It does not eliminate the need for the test.

Sources

• Z.ai, “GLM-Legal: Advancing Agentic Legal Intelligence”
• Artificial Lawyer, “Chinese Company Launches Legal-Specific LLM”

USA Today Co. Sues OpenAI Over Alleged Use of Newspaper Content in AI Training

The new Southern District of New York complaint seeks more than $250 million and asks for destruction of models and training sets alleged to incorporate plaintiffs’ work. The claims have not been adjudicated.

USA TODAY Co., Inc. and 13 affiliated plaintiff entities filed a copyright complaint against seven OpenAI defendants on October 8, 2026, in the U.S. District Court for the Southern District of New York.

The case, USA TODAY Co., Inc. v. OpenAI Foundation, No. 1:26-cv-08892, is another publisher-side challenge to the use of news content in generative AI systems. But it is important to keep the posture straight: this is a newly filed complaint, not a court ruling. The allegations have not been tested in court, and no merits decision has been verified.

According to the complaint, the 14 plaintiffs say they own copyrights in content published by 19 publications and allege that OpenAI acquired and used that content without authorization in model training and outputs. The complaint asserts three counts: copyright infringement, vicarious copyright infringement, and removal of copyright-management information under 17 U.S.C. § 1202. Its willfulness and damages assertions are also allegations.

Those are the plaintiffs’ allegations. The filing does not establish that OpenAI committed infringement, removed copyright-management information, acted willfully, or owes damages.

A demand exceeding $250 million and a request for model destruction

The plaintiffs seek damages exceeding $250 million, according to the complaint. In the prayer for relief, they separately ask the court to permanently enjoin the alleged conduct and to order, under 17 U.S.C. § 503(b), destruction of all GPT or other LLM models and training sets that incorporate their content.

Those are requested remedies, not relief the court has granted. Whether the court awards damages, enters an injunction, or orders destruction will depend on later proceedings and rulings that have not occurred.

This complaint adds another publisher action to the pending disputes over the use of copyrighted journalism in generative-AI development. OpenAI has not yet responded in this case, and the reviewed docket does not establish which defenses it will assert. No claim or defense has been resolved here.

The case may join a broader OpenAI copyright litigation cluster—but has not yet done so

The plaintiffs filed a statement of relatedness asking that the case be treated as related to In re OpenAI, Inc. Copyright Infringement Litigation, No. 1:25-md-03143.

If the court deems the case related or brings it into the MDL, that could affect assignment, scheduling, discovery, and coordination with overlapping cases. A public RECAP docket snapshot reviewed at 7:25 a.m. ET on October 9, 2026, did not show an order accepting relatedness, consolidating or reassigning the case, deciding the merits, or a response from OpenAI. Because no PACER account is available for this package, that snapshot is the available public record rather than an authenticated PACER refresh. The case therefore remained a newly filed complaint in the Southern District of New York on the record reviewed.

Why the filing matters

The lawsuit is a timely marker of the pressure surrounding training-data practices, particularly where the underlying works are commercially produced journalism. Its significance lies less in any immediate legal holding than in the scope of the allegations, the number of publications involved, and the breadth of the relief requested.

For publishers, the filing reflects a continuing effort to define the legal boundaries of using news content in generative AI systems. For AI companies, it is another case that may test how copyright law applies to model training and outputs.

Potential next developments include court action on relatedness or consolidation, OpenAI’s response, and later rulings on the pleaded claims. None should be assumed from the complaint alone.

Public docket snapshot: CourtListener RECAP docket, No. 1:26-cv-08892 Official docket: PACER Case Locator (search S.D.N.Y. No. 1:26-cv-08892) Filed complaint copy: DocumentCloud mirror of Dkt. 1