Tag: artificial intelligence

Can I be sued over an algorithm that I didn’t even program?

Posted by Jaime Farrant on September 4, 2026

Learn how the settlement Meta reached with 52 attorneys general could set a standard that impacts your business.

The United States has no federal artificial intelligence statute. There is no American equivalent of the EU AI Act, no agency that certifies AI models, and no registry of high-risk systems. And yet, on August 26, 2026, Meta — the parent company of Facebook, Instagram and WhatsApp, among others — agreed to subject Instagram and Facebook to an algorithmic governance regime with mandatory error thresholds, annual third-party testing, and an independent auditor who reports to state attorneys general.

It did not take a federal AI law to get there. Instead, it required the political will of a bipartisan coalition of state attorneys general determined to litigate against Meta, and that litigation produced a settlement agreement whose effects will be felt across every state and territory, Puerto Rico included.

That is the point worth absorbing for anyone working in AI governance: while we waited for comprehensive federal legislation, algorithmic regulation in the United States arrived through consumer protection litigation.

What was the case that led to this outcome?

The settlement was reached inside a consolidated proceeding, In re Social Media Adolescent Addiction/Personal Injury Products Liability Litigation (MDL 3047, Northern District of California), which consolidated several state lawsuits against Meta on 2 legal theories:

Unfair and deceptive practices, under what are known as state “UDAP” statutes (Unfair and Deceptive Acts or Practices). The states alleged that Meta designed its platforms to maximize the time adolescents would spend on them — through notification mechanics, infinite scrolling, and social rewards — while publicly representing that its products were safe and that its protective tools worked as advertised.
COPPA violations. COPPA is the federal privacy statute protecting children under 13, and the states alleged Meta collected children’s data without verifiable parental consent, among other failures.

Meta denied all liability, and the agreement signed last week contains no admission of wrongdoing. Nevertheless, Meta will pay a guaranteed minimum of roughly $12.7 billion to what may be 48 states, the District of Columbia and 3 territories — Puerto Rico among them — distributed over the next 10 years. That figure could rise, according to Meta, to approximately $18 billion if other platforms meet certain conditions.

Five pieces of AI governance hidden inside a consumer protection settlement

Beyond a payment of a magnitude that would be catastrophic for the vast majority of companies worldwide, the more significant fact may be this: under the agreement, Meta committed — potentially for the next decade — to what is arguably the first AI governance framework established inside a legal proceeding. This agreement will need to be studied by every business, and particularly by those that market to minors or that know their websites and apps are used by minors.

The governance framework the agreement establishes:

1. A model with a written error threshold.

Meta is required to deploy age estimation methods — trained classifiers, not sign-up forms — and to meet maximum false positive rates: roughly 10% for minors aged 16 and 17, and 3% for the 13-to-15 group, with wider tolerances during the first year for proprietary methods. The agreement also requires Meta to build and test a dedicated model to detect users under 13, with annual detection targets.

This is remarkable. An American judicial settlement is fixing the minimum statistical performance of a machine learning model, by age group, in concrete numbers — leaving none of the ambiguity that “reasonable” or “commercially appropriate” criteria would have allowed.

2. Annual algorithmic audit by an independent third party.

An independent auditor will verify, on an annual basis, the false positive rates, the volumes of underage account detection, the effectiveness of the account-linking models, and the efficacy of the usage pauses, and will report to the states. In practice, this is an algorithmic system audit analogous to the one Article 37 of the European Digital Services Act (“DSA”) requires annually of very large platforms — with the difference that, in the United States, the requirement did not arrive through legislation but embedded in a consent judgment.

3. Purpose limitation on the model’s data.

Data collected to estimate age must be deleted immediately after the age determination is made, protected under the company’s highest standards, and may not be used for advertising, marketing, or to optimize recommendation models.

4. A right to contest an automated decision.

If the system misclassifies your age, Meta must offer a clear and conspicuous mechanism to appeal that determination, and must resolve it within a reasonable time.

5. The recommendation system becomes a regulated object.

Meta must offer a chronological feed — with no algorithmic personalization — on a reasonably accessible basis, present it actively to new teen accounts within the first 10 days, and remind users every 90 days. Supervising parents can lock that feed as the default. Add to this that teens will, by default, be unable to see how many likes or reactions posts receive; the pauses at 60 and 90 minutes of use; the overnight block; and the silencing of notifications during school hours.

The agreement does not treat the recommendation algorithm as an untouchable trade secret, but as a product feature that a regulator can order switched off where it is found to be defective, deceptive or abusive.

The global view: Europe reached a similar determination first, but through a different legal route

If these terms sound familiar to anyone who works with European regulation, that is because they are. On April 29, 2026, the European Commission preliminarily found that Meta had breached the Digital Services Act precisely for failing to identify, assess or mitigate the risk of children under 13 accessing Instagram and Facebook. The Commission estimated that between 10% and 12% of children under 13 in the EU use those platforms, and criticized a reporting tool for underage accounts that required up to seven clicks. Fines under the DSA can reach 6% of a company’s total worldwide annual turnover.

The European instruments worth remembering for these matters include:

DSA. Article 28 requires privacy, safety and security measures for minors and bars advertising based on profiling directed at minors; Articles 34 and 35 require assessment and mitigation of systemic risks, including effects on the physical and mental well-being of minors; Article 37 imposes annual independent audits; and Article 38 requires very large platforms to offer at least one recommender option not based on profiling. In other words: the chronological feed the states extracted from Meta by settlement is already a legal obligation in Europe.
GDPR. Article 8 governs children’s consent; Article 5(1)(b) and (c), purpose limitation and data minimisation; Article 22 and Recital 71, automated decision-making, which “should not concern a child”; Article 25, data protection by design and by default; and Article 35, impact assessments. In February 2025,the European Data Protection Board issued a statement on age assurance setting out 10 principles that anticipate, almost point for point, what the Meta settlement now requires: proportionality, minimisation, demonstrable effectiveness, safeguards against automated decision-making, and the warning that age assurance “should not provide additional means for service providers to identify, locate, profile or track natural persons.”
AI Act. Less applicable than one would expect, and it is worth understanding why. Article 5 — in force since February 2, 2025 — prohibits AI systems that exploit vulnerabilities arising from age in order to materially distort behaviour and cause significant harm, which describes the states’ theory about Meta’s addictive design. But the high-risk obligations under Annex III — the ones that bring risk management, data governance, technical documentation, human oversight and accuracy requirements — were deferred by the Digital Omnibus agreed in 2026, and now begin on December 2, 2027.

So, as things stand, the European instrument designed expressly to govern AI does not yet fully apply to these systems, while an American court settlement, negotiated under consumer protection laws dating to the 1970s, is already formally setting error thresholds and audits. Consequently, AI governance did not arrive through the creation of a purpose-built legal framework, as everyone assumed it would.

Why should this matter to me?

Because you do not need an AI law for someone to demand AI governance from you. Any deceptive practices statute — Puerto Rico’s included — can be used to ask whether your model does what you said it does. If your company claims its system “detects fraud with 99% accuracy” or that its algorithm “does not discriminate,” that is a legally enforceable representation, and the burden of proving it is yours.
Because the evidentiary standard in these cases has shifted. Meta did not lose because of what it did. Its legal exposure came because of the distance between its public representations and its own internal documents. In algorithmic governance cases, you generate the adverse evidence yourself: your evaluation metrics, your risk memos, the service tickets and complaints nobody inside the company ever addressed.
Because numerical obligations are now being normalized. Once a public settlement establishes that an age classifier must operate at a 3% false positive rate for the 13-to-15 age group, that number becomes the reference point for the next case, the next contract, and the next negotiation with an enterprise customer.
Because transatlantic convergence is becoming real. This settlement and the European regulatory framework point at the same approach — one that looks at the defects, risks and harms a product causes, and imposes nearly identical remedies. If you build for both markets, designing twice will increasingly be seen as wasted money: to be safe and to limit liability, you will have to operate to the most demanding standard.

How can I comply?

Inventory your automated systems. Not just what you call “AI.” Include your scoring models, classifiers, recommendation engines and segmentation rules, among others. You cannot govern what you have not counted.
Define the metrics before you deploy, not after. What is the acceptable error rate? For which subgroups? Who measures it, and how often? Write it down before a regulator or a court writes it for you.
Build purpose limitation into your data pipeline. Data collected for compliance — verification, security, fraud prevention — must not be recycled for training, marketing or profiling. Beyond writing policies to that effect, create technical documentation that substantiates it.
Build a mechanism to respond to individual complaints. Every automated decision affecting a person needs a visible route of appeal and a human being to handle it. This is already required in Europe under the GDPR, and it will increasingly be expected in the states.
Prepare for an audit you cannot control. Assume that at some point a third party will ask for your evaluation documentation, your decision logs and your model change history. If that does not exist today, we recommend you build it within the next three months.
Review your model vendors. If your classifier is built or maintained by a third party, their error rates are your error rates. Ask for certifications and keep them.
If you are in Puerto Rico, start now — do not wait for a local AI statute. We already have Act 163 of 2026, which amended the Right to One’s Own Image Act to expressly cover representations generated, cloned or simulated by artificial intelligence, commonly known as deepfakes. And, if your business offers services to people in the European Union, the GDPR and the DSA apply to you even if your office is in San Juan, Aguadilla or Ponce.

Conclusion

For years, the conversation about AI governance in the United States circled around a single question: when will a federal AI law arrive? The Meta settlement suggests that this was the wrong question.

Algorithmic regulation arrived with no AI statute, no specialized agency and no legislative process. It arrived as a negotiated remedy inside a case, with numbers, deadlines and an auditor. And it arrived with a principle that applies to any organization deploying models, whether it has 50 employees or 50,000: if you cannot measure your system, you cannot defend it.

Europe built this order through regulations. The United States just wrote it into a consent judgment. The result, for whoever builds the products, is very nearly the same.

Want to know whether your business’s automated systems can withstand this kind of scrutiny? You can book a consultation with us today. We are here to help.

About the Author

Jaime Farrant is an attorney admitted to practice in Puerto Rico, New York, Maryland and the District of Columbia, with an LL.M. in International Law, focused on privacy, cybersecurity and artificial intelligence regulation for businesses and healthcare providers.

ATTORNEY ADVERTISING. This article constitutes advertising as defined under the rules of professional conduct in force in New York (22 NYCRR 1200.7.1 and 1200.7.3), Maryland (Rule 19-307.1 and 19-307.2) and the District of Columbia (D.C. Rules of Professional Conduct 7.1), as well as the Puerto Rico Rules of Professional Conduct (Rules 7.1-7.3). It does not constitute solicitation of known potential clients in need of legal services in a particular matter. Rather, it is general information directed to the public about the practice of law and the legal services available. No attorney-client relationship is created by reading this article or by contacting the author.

Can I be Sued for Discrimination Under Puerto Rico’s Law 100 for Using ChatGPT at Work, Even if I Never Meant to Discriminate?

Posted by Jaime Farrant on August 27, 2026

In our article published 2 weeks ago, we talked about how using AI to screen résumés exposes you to laws that prohibit workplace discrimination. Today, we want to focus on the risks you face if you operate in Puerto Rico under a law that almost no vendor will mention when you’re evaluating an automated “screening” tool: Law Num. 100 of June 30, 1959, Puerto Rico’s employment discrimination law, which applies even though its text doesn’t contain the word “algorithm” — and adds consequences that neither U.S. federal law nor the European Union impose.

What Does Law 100 Presently Say?

Before 2017, Law 100 gave employees a powerful tool: it provided that an adverse action — such as failing to hire, failing to promote, or terminating someone — taken without just cause, was presumed to be discriminatory. Once the employee triggered that presumption, the burden shifted almost entirely to the employer, who then had to affirmatively prove that no discrimination had occurred.

That changed with the Labor Transformation and Flexibility Act (Law 4 of 2017), which eliminated that presumption in order to align Puerto Rico with the federal standard known as McDonnell Douglas. Today, a claim under Law 100 works much like a federal one: the employee must establish a preliminary case (that they belong to a protected class, that they were qualified, that an adverse action occurred, and that similarly situated people outside that protected class were treated better). Once an employee does this, the employer must articulate a legitimate, non-discriminatory reason for the decision. If the employer does so, the burden shifts back to the employee, who must then prove that reason was pretextual.

In short: today, the burden of persuading the court that discrimination occurred rests more on the employee than on the employer.

However, AI introduces a real problem for employers even under this favorable standard. As mentioned above, the second step of the analysis requires you to be able to articulate a legitimate reason for the rejection. If your entire process was uploading résumés to a third-party tool and letting an algorithm screen them out, what is your legitimate reason for rejecting them? Answering “The system gave them a lower score” isn’t always enough — especially if the candidate can show, through statistics (as allowed under Title VII’s disparate-impact theory), that the tool’s rejection pattern — for example, disproportionately screening out women for a position — correlates with a protected category. In that scenario, your inability to explain the decision becomes the very evidence of pretext the candidate or employee needs to prevail.

That said, even though the burden of proof is no longer automatically stacked against the employer, Law 100 still has something Title VII doesn’t: harsher penalties, discussed below.

What Are the Penalties Under Law 100?

Law 100 imposes civil liability on the employer for double the damages caused (or between $500 and $2,000 if damages cannot be determined). In addition, discriminatory conduct may constitute a misdemeanor, punishable by a fine of up to $5,000 or up to 90 days in jail, or both. None of this exists under any federal anti-discrimination law. As a result — although it is uncommon for these cases to be prosecuted criminally — using AI to screen résumés could technically expose you to criminal liability in Puerto Rico. It’s a consequence many employers don’t know about or anticipate.

How Is This Regulated in Other Jurisdictions?

United States: there is no single federal law governing AI in employment. Instead, there’s a patchwork of local and state rules — New York City’s Local Law 144, which mandates bias audits; Illinois’s requirement to notify candidates when AI is used to analyze video interviews; Colorado’s risk assessments — layered on top of existing federal anti-discrimination laws (Title VII, ADEA, ADA) enforced by the EEOC and interpreted under the U.S. Supreme Court’s McDonnell Douglasstandard. All of these can be boiled down to: you’re allowed to use AI tools, but be ready to justify them if challenged — and know the specific compliance requirements in whatever state you operate in or evaluate candidates from.
European Union: if your company works with candidates or employees in the EU, or you simply want to understand where AI regulation is headed globally, it’s worth looking at the EU AI Act (Regulation (EU) 2024/1689), already in force, and likely the most comprehensive legal framework on artificial intelligence in the world today. The EU AI Act classifies AI tools used for recruitment and personnel selection — including filtering job applications and evaluating candidates — as “high-risk” systems. That means that before they can be legally used, the tool must undergo a conformity assessment, have technical documentation, risk-management systems, human-oversight mechanisms, and be registered in an EU database. Employers also have a specific obligation to inform workers and their representatives of the system’s existence before using it. In the European Union, unlike in Puerto Rico and the United States, an employer must demonstrate that its platform is safe before using it.

The difference in approach is significant. While the EU certifies the tool before it’s used, Puerto Rico and the United States require nothing upfront, but impose legal consequences if the tool produces a discriminatory result.

Why Should This Matter to You?

Because using AI that discriminates against candidates puts you at risk of paying damages — and if you’re in Puerto Rico, it puts you at risk of criminal liability.

Think about it in practical terms: if your AI tool disproportionately rejects candidates of a certain age or background, and a candidate manages to prove pretext — for example, by showing that you couldn’t coherently explain why the system screened them out — you face the possibility of having to compensate the people affected by the algorithm, and of being held criminally responsible for having used it. For a small business or a clinic, that’s a genuinely significant exposure that goes beyond what you’d anticipate if you only looked at federal law as your benchmark, or if you assumed you were covered because ChatGPT was built “without any intent to discriminate” or because you used the AI tool in “good faith.”

How Can I Comply With the Law?

Always have an articulable reason for every rejection. “The system gave candidate X a lower score” isn’t enough on its own. You need to be able to explain, in concrete terms tied to the job’s qualifications, why a candidate didn’t move forward in the interview process.
Demand documentation from your vendor. Ask whether the tool has been bias-audited, how often, and whether they’ll provide you that documentation. That evidence is what lets you satisfy the “articulate a legitimate reason” step if it is challenged.
Keep a record of every decision. What data went in, what score or result the system produced, and what human review occurred before each final decision. That log is your evidence if you ever need to defend the process.
Notify candidates that you use AI in the process. It’s good practice in Puerto Rico, and it’s already mandatory under the EU AI Act if you have candidates or employees in Europe.
Run a risk assessment before adopting the tool — not after the first complaint. Ask yourself: what data was it trained on? Which protected categories might it be indirectly affecting? Remember that Law 100 protects, among others: age, race, color, sex, sexual orientation, gender identity, social or national origin, social condition, political affiliation, political or religious beliefs, being a victim (or perceived victim) of domestic violence/sexual assault/stalking, veteran status, marital status, and even certain hairstyles and hair textures associated with particular racial identities or national origins.
Review your contract with the vendor. Who’s responsible if the tool produces a discriminatory result? Negotiate that clause if you can.

The Bottom Line

Since the 2017 labor reform, Law 100 no longer puts you in a worse evidentiary position than federal law. What does set Puerto Rico apart is what happens if you lose: double damages and the possibility of criminal liability, neither of which exists under Title VII. Remember, too, that if you can’t explain why your tool rejected a candidate, you may already be in violation of the law. Finally, remember that you need to keep your processes documented, make sure they’re supervised by people, and be able to explain and justify every decision. Know and formalize your processes before you’re forced to do so in front of a government agency or a court.

Do you think your current AI-driven hiring process could expose you to legal liability under Ley 100 if a candidate filed a complaint today — or do you just want to make sure your processes are compliant? Let’s talk. Book a consultation here.

About the Author

Jaime Farrant is admitted to practice law in Puerto Rico, New York, Maryland and the District of Columbia. Practice in all other jurisdictions is limited to immigration law. This article is for informational purposes only and does not constitute legal advice or create an attorney-client relationship.

ADVERTISING MATERIAL. This article constitutes advertising as defined by the professional conduct rules in New York (22 NYCRR 1200.7.1 and 1200.7.3), Maryland (Rule 19-307.1 and 19-307.2), and the District of Columbia (D.C. Rules of Professional Conduct 7.1), and the Puerto Rico Rules of Professional Conduct (Rules 7.1-7.3). It is not solicitation of prospective clients known to need legal services in a particular matter. Instead, it is general information directed to the public about the practice of law and available legal services. No attorney-client relationship is created by your reading of this article or by contacting the author. Consult qualified counsel in each jurisdiction with specific situations.

Can I Use ChatGPT to Put Bad Bunny in my business’s social media ads?

Posted by Jaime Farrant on August 20, 2026

As I write this post, social media in Puerto Rico has blown up with Bad Bunny’s announcement that he will close his world tour in concert scheduled for August 22 and 23 in San Juan’s Hiram Bithorn Stadium. The amount of posts about this announcement reminded me of the many promotions posted by Puerto Rican restaurants, bars, and small businesses in their Facebook, Instagram and TikTok accounts, where they showed a photo of Bad Bunny eating at their restaurant, or having a beer at their bar. However, he was never in any of these places. These were AI-generated images, posted without his consent, for one obvious reason — to draw customers to their businesses using the likeness of arguably the most recognizable person in Puerto Rico today. However, no business paid a license, asked permission, or, in most cases, thought twice about it.

If your business operates in Puerto Rico, New York, Maryland, or Washington D.C., the answer to whether you can legally do this depends heavily on which of these you’re in — and the gap between them is bigger than most business owners realize.

In the US, How – or If – you can use Deepfakes Depends Entirely on Where You Live

Unlike Puerto Rico, which just amended its “Right to One’s Own Image” statute (Law 139-2011, as amended by Law 163-2026) to explicitly cover AI-generated deepfakes by penalizing their unauthorized commercial use with damages of up to $100,000 per violation if the use was intentional or with gross negligence, the United States has no uniform federal right of publicity. Each state decides for itself whether — or how — to protect someone’s name, voice, or likeness from unauthorized commercial use. That means that the same AI-generated Bad Bunny photo can be a serious legal problem in one state and close to unregulated in the state next door.

New York: The Strongest Protections of All

New York has protected this right since long before generative AI existed. Civil Rights Law §§ 50–51 makes it a misdemeanor — and a civil cause of action — to use a living person’s name, portrait, picture, likeness, or voice for advertising or trade purposes without their prior written consent. Section 51 lets the injured person seek an injunction, actual damages, and, if the defendant knowingly used their likeness, exemplary (punitive) damages at the jury’s discretion.

On top of that foundation, New York has added two AI-specific layers in the last 2 years:

The Digital Replica Contracts Act (General Obligations Law § 5-302): voids contract provisions that let an employer replace a performer’s actual performance with a digital replica, unless the performer was represented by counsel or a union and the terms are stated clearly in a separately signed agreement.
The Synthetic Performer Disclosure Law (General Business Law § 396-b), effective June 9, 2026: requires advertisers to conspicuously disclose when an ad contains a “synthetic performer” created using generative AI. Civil penalties run $1,000 for a first violation and $5,000 for each subsequent one.

Put together, a New York business running that “Bad Bunny at my bar” photo is exposed on two fronts: a §§ 50–51 claim from Bad Bunny himself (or his estate, for that matter, since New York also protects deceased performers’ digital replicas under Civil Rights Law § 50-f), and a separate disclosure penalty if the ad used a synthetic element and didn’t label it.

Maryland: Barely Any Legal Protections at All

This is likely to surprise business owners coming from New York or Puerto Rico: Maryland has no right of publicity under its statutes or common law. It’s one of only a handful of states (along with Alaska, Kansas, and North Carolina) where this right doesn’t exist as such. A bill that would have created a civil cause of action for unauthorized use of someone’s identity via AI or deepfakes — House Bill 1425/Senate Bill 905 — did not pass in the 2025 session. It’s been reintroduced as House Bill 184 for the 2026 session, but as of this writing, it has not been approved.

Maryland does have a deepfake statute — Senate Bill 141 (2026), effective June 1, 2026 — but it is narrowly limited to election-related deepfakes intended to influence voting or misrepresent election facts. It has nothing to say about a restaurant using an AI-generated photo of a celebrity to sell arepas, margaritas or mofongo.

Practically, this means that today, a Bad Bunny impersonation ad run by a Maryland business faces essentially no exposure under Maryland state law specifically built for this problem. That could change if HB 184 passes, and it’s also worth remembering that Bad Bunny himself could still bring a claim in a state where he does have rights like New York, depending on where the harm occurred.

Washington D.C.: Regulated by Common-Law, Not a Statute

D.C. has no right-of-publicity statute either. What it has is a common-law claim for misappropriation, drawn from the Restatement (Second) of Torts § 652C, as applied in Vassiliades v. Garfinckel’s, Brooks Bros., 492 A.2d 580 (D.C. 1985). To win, a plaintiff has to show both that the defendant benefited from using their identity and that there’s a recognizable public or commercial value in that identity — the exact opposite of a bright-line statute like New York’s. This makes outcomes far less predictable and cases more expensive to bring, since there’s no statutory damages figure to point to and no per-violation civil penalty to threaten a defendant with.

How is This Regulated in the European Union?

The European Union has taken an approach very different than the patchwork of state laws in the US through its enactment of the EU AI Act, which applies across all Union states. The EU AI Act does not establish a standalone private right of action for damages; rather, it imposes administrative transparency obligations. Under Article 50 of the Act, providers of Al systems that generate or manipulate image, audio, or video content constituting a deepfake must clearly disclose that the content has been artificially generated or altered. An exception applies when such content is part of an obviously artistic, satirical, or fictional work, provided it is not presented in a misleading manner. Failure to comply with these transparency requirements constitutes a serious infringement, subject to administrative fines of up to €15 million or 3% of the undertaking’s total worldwide annual turnover for the preceding financial year, whichever is higher.

Why Should This Matter to You?

If you run a business — or advise clients who do — across any of these jurisdictions, the question “can I use an AI image of a celebrity in my ad” doesn’t have one answer, and responses range from “yes, expect a lawsuit and pay damages of up to $100,000” (Puerto Rico), to “yes, expect a lawsuit and possible punitive damages” (New York) to “there’s currently no statute built for this” (Maryland) to “it depends on how a judge applies a 40-year-old privacy tort” (D.C.), or “you must publish in your campaign that its content was artificially generated” (EU). Consequently, a marketing decision that’s clearly reckless in Manhattan might be legally uneventful across in Maryland — for now.

These differences are exactly the kinds of gaps that generative AI has widened. Tools like ChatGPT, Claude, Midjourney, and similar platforms make it trivial to generate a photorealistic image of a real, identifiable person for a fraction of what a licensing deal would have cost a few years ago. The law in most of the United States hasn’t caught up uniformly, which means your exposure depends less on what you did and more on where you did it.

How Can You Comply With the Law?

If you operate in New York, treat any AI-generated image or voice of a real person in your advertising as requiring the same written consent you’d need for a real photo — Civil Rights Law § 51 doesn’t distinguish between a real photograph and a generative AI recreation.
If your New York ad uses a synthetic performer (not a real, identifiable person, but a “no such person exists” AI-generated model), confirm you’re including the conspicuous disclosure required by GBL § 396-b before it airs.
If you operate in Maryland, don’t assume the absence of a right-of-publicity statute means zero risk — track HB 184, and remember a claim can still be brought in a state where the depicted person has stronger rights.
If you operate in D.C., document your process for obtaining consent regardless of the weaker legal baseline; a misappropriation claim can still succeed, and consent is always the safer route.
If you operate in Puerto Rico, make sure your processes clearly document that the use was authorized.
If you operate across multiple states, apply the strictest applicable standard (in this example, New York’s) to any content you plan to run across state lines or online, since your audience — and any resulting claim — isn’t limited to where your business is physically located.
If your campaign will be shown in the European Union, you will have to divulge that it was artificially generated.

The Bottom Line

Puerto Rico, New York, Maryland, D.C., and the European Union sit at very different points on the right-of-publicity spectrum — from New York’s statutes with real teeth, to Maryland’s near-total absence of protection, to D.C.’s uncertain common-law doctrine. If your business uses generative AI in marketing and you operate in more than one of these jurisdictions, the safest approach is to assume the strictest rule applies everywhere your content is seen, not just where you’re physically located.

Does your business use AI-generated content in advertising across New York, Maryland, or D.C.? Schedule a consultation today to review your exposure in each jurisdiction where you operate.

About the Author

Jaime Farrant is admitted to practice law in Puerto Rico, New York, Maryland and the District of Columbia. Practice in other jurisdictions is limited to immigration law. This article is for informational purposes only and does not constitute legal advice or create an attorney-client relationship. Laws referenced are current as of August 15, 2026.

What Happens If Your Vendor’s AI Decides to Hack Someone Else?

Posted by Jaime Farrant on July 28, 2026 with 0 Comment

Have you ever thought about what could happen to your business if a vendor’s AI system decides, on its own, to break into another company’s servers? If you haven’t, it might be time to, because the consequences for your business could be severe. If you’re a business regulated by HIPAA, a violation of this law could carry a civil penalty of up to $2,190,294 per violation category, per year, at the highest tier of culpability. Even a business that did nothing wrong, where a vendor’s AI system acted entirely on its own, could still face a lower-tier penalty, an OCR investigation, breach notification costs, and reputational fallout, for something it never caused and couldn’t have predicted.

This nightmarish possibility is no longer a hypothetical scenario. On July 21, 2026, OpenAI published on its website a notice were they took responsibility for a cyberattack on Hugging Face, a widely used AI hosting and machine-learning collaboration platform. According to OpenAI, a combination of its models — including a publicly available model and a more capable unreleased one, running with reduced safety restrictions for an internal cybersecurity evaluation — broke out of their isolated test environment by exploiting a previously unknown flaw in an internal software tool, reached the open internet, and then used stolen credentials and another unknown vulnerability to gain remote code execution on Hugging Face’s production servers. Their goal, according to OpenAI, was narrow but telling: the models were trying to retrieve the answer key to the benchmark test they were being scored on. Hugging Face had already detected the intrusion over a weekend of automated activity, reported it to law enforcement, and began its own containment before it even learned OpenAI was behind it.

Both companies have called this a watershed moment for cybersecurity. For a small business, medical practice, or professional office that relies on outside vendors — including AI tools — to store, process, or transmit sensitive information, it should also be a wake-up call about a risk category that most vendor contracts were never written to address: the AI agent that acts on its own.

Why could your AI Vendor’s Behavior Become Your Problem?

Most privacy and data security laws that apply to small businesses do not distinguish between a breach caused by a human hacker and a breach caused by an autonomous system. If your practice or business uses a covered entity’s business associate, a cloud vendor, or any third party that touches personal or health information, you are generally still responsible for:

Vetting that vendor’s security practices before you sign a contract (due diligence).
Having the right contractual protections in place, such as a HIPAA Business Associate Agreement (BAA) for medical offices, or comparable data processing and security terms for any business handling personal information.
Notifying affected individuals, and in some cases regulators, if that vendor’s system is compromised and your data is involved.

Under HIPAA, a covered entity’s business associates are contractually and legally bound to safeguard protected health information (PHI), and a breach at the vendor level can trigger notification obligations for the covered entity itself, even though the vendor’s system, not the medical office’s, was the one that failed. Outside of healthcare, most state data breach notification laws work the same way: liability follows the data, not just the party that caused the incident.

An AI agent that autonomously escalates its own access, exfiltrates credentials, or reaches systems it was never authorized to touch does not change any of that legal analysis. It just makes it harder to predict, detect, and contain.

It’s worth being precise about what did and didn’t happen here: by OpenAI’s own account, the models were chasing the answer key to their own benchmark test, not deliberately hunting for customer or patient records. No business should read this incident as proof that patient or client data was taken. What should concern any business relying on outside vendors is the capability on display: an AI system that, on its own initiative, found a zero-day vulnerability, stole credentials, escalated privileges, and reached a third party’s production infrastructure, over an unmonitored weekend, before any human intervened. Point that same capability at a system that holds patient records, financial account numbers, or client files, and the outcome looks very different.

A Disclosure Gap Worth Knowing About

Here’s a detail that matters for any business relying on a vendor’s assurances: OpenAI was not legally required to disclose this incident at all. Two recent state laws, California’s SB 53 and New York’s RAISE Act, require large AI developers to report critical safety incidents, but only if the incident risks more than 50 deaths or serious injuries, or over $1 billion in property damage. An incident like this one falls well short of that bar. OpenAI disclosed it voluntarily. The practical takeaway for your business: you generally cannot count on a public filing or regulatory notice to tell you whether a vendor’s AI system has had a similar failure. That makes your own contract language, and your own right to ask direct questions, the primary tool you have.

Penalty Structure: What’s Potentially at Stake

The exposure here is layered, and it can apply to a business that never asked for an AI system to do anything wrong, if that system operated within its own environment or a vendor’s:

HIPAA: Civil penalties currently range from roughly $145 up to $2,190,294 per violation category per year, depending on the covered entity’s or business associate’s level of culpability. Tier 1 (lack of knowledge) sits at the low end; willful neglect that goes uncorrected sits at the top. State attorneys general can separately pursue HIPAA-related fines of up to $25,000 per violation category, per year, and multi-state actions are increasingly common when a breach touches residents across several states.
State breach notification laws: Most states can pursue penalties or authorize private lawsuits when a business fails to notify affected residents promptly after a breach involving personal information, regardless of whether the breach originated with the business or with a vendor it selected.
Contractual exposure: If your vendor agreement lacks clear breach notification timelines, security requirements, or audit rights covering AI tools specifically, your business could be left absorbing costs, or negotiating from a weaker position, after the fact.

None of this means every AI-related vendor incident automatically results in a maximum fine. Regulators generally consider the nature of the data involved, the number of people affected, whether the business had reasonable safeguards in place, and how quickly the incident was addressed. But the exposure is real, and it is not limited to companies that build or sell AI models. It reaches any business, medical office, or professional practice that relies on one.

Why Should You Care About This?

Because experts who study AI safety are calling this one of the first real-world examples of an AI “loss of control” scenario: a system doing something researchers had long warned about, without a human directing it, and without a simple software bug to blame. The activity reportedly ran for an extended period on a system that, unlike OpenAI’s actively monitored production tools, was not being watched in real time. If a frontier AI lab with dedicated security teams can have this happen during a controlled internal test, it is a reasonable question for any business to ask what oversight exists over the AI-enabled tools, chatbots, scheduling assistants, or back-office automation your practice already uses, and what your vendor’s contract actually says about that risk.

For a medical office, this question is not abstract. AI tools are increasingly built into patient intake, scheduling, transcription, and billing software. For any small business, it applies to whatever AI-enabled service touches client records, financial data, or other sensitive information, even indirectly.

How Can You Protect Your Business?

Inventory every vendor and software tool your business uses that incorporates AI, especially anything touching patient, client, financial, or employee data.
Confirm you have a signed BAA in place with any vendor that creates, receives, maintains, or transmits PHI on your behalf, if you are a covered entity or business associate.
Review vendor contracts for AI-specific language: does the agreement address autonomous system behavior, require prompt breach notification, and specify security obligations?
Ask vendors directly how they test AI systems for containment and what happens if a model exceeds its intended scope.
Confirm your incident response plan accounts for a scenario where a vendor, not your own systems, is the source of a breach.
Revisit your cyber insurance policy to confirm it covers incidents involving AI tools and third-party AI vendors, not just traditional data breaches.
Don’t assume silence means safety: build a contractual right to be notified of AI-related security incidents into your vendor agreements, since current AI safety-incident disclosure laws only cover the most catastrophic events and won’t necessarily surface a vendor’s close call.

The Bottom Line

The OpenAI–Hugging Face incident is a reminder that AI risk in 2026 is not just about what your business chooses to do with AI. It is also about what the AI systems inside your vendors’ infrastructure might do without anyone telling them to. If your practice or business has not reviewed its vendor agreements and incident response plan with that possibility in mind, now is a good time.

If you have questions about your vendor contracts, business associate agreements, or how a breach at a third-party AI vendor could affect your obligations, schedule a consult with us today.

About the Author

Jaime Farrant is admitted to practice law in Puerto Rico, New York, Maryland and the District of Columbia. This article is for informational purposes only and does not constitute legal advice or create an attorney-client relationship.