Richmond County Courthouse sign in Augusta, Georgia. Shutterstock

Richmond County Courthouse sign in Augusta, Georgia. 

Why Public Defense Should Incorporate AI Carefully

A school bus in in Marilia, São Paulo, Brazil, April 12, 2025. Shutterstock

A school bus in in Marilia, São Paulo, Brazil, April 12, 2025. 

Brazil's Data Protection Agency Faces Landmark Test on Kids and Facial Recognition

People visit the booth of Kimi, an LLM developed by the Chinese startup Moonshot, during the World AI Conference in Shanghai, China, Monday, July 20, 2026. (FeatureChina via AP Images)

US Government’s AI Risk Review Should Apply to Open Weight Models

Baden-Württemberg, Stuttgart: A woman is holding a pair of smart glasses from Ray-Ban Meta in her hand at an eyewear store. (Christoph Schmidt/picture-alliance/dpa/AP Images)

For Meta’s Smart Glasses, Access Is An Afterthought

A recording of Meta Founder and CEO Mark Zuckerberg's deposition is played for the jurors on March 4, 2026, in Santa Fe, N.M. (Jim Weber/Santa Fe New Mexican via AP, Pool, File)

Courts Find Platforms Addictive and Dangerous by Design

Meta CEO Mark Zuckerberg attends UFC Freedom 250 on the South Lawn of the White House, Sunday, June 14, 2026, in Washington. (AP Photo/Alex Brandon)

No Zuckerberg, The Future is Not Just For Individuals

A protestor carries a poster with a message in Hindi which translates as 'Save Our Constitution' as he participates in a protest movement led by Cockroach Janta Party, in New Delhi, India, Wednesday, July 22, 2026. (AP Photo/Vipin)

BitChat-Github Takedown Shows India Redrawing Its Perimeter of Online Control

A collage made of 13 different images arranged in a rectangular formation in the style of pop art. The images include vintage photographs of old signs, part of a vintage poster, a vintage space image, historic black and white photos of a male laborer carrying a sack and of a human eye, comic book-style words 'Wow!', 'Bam!' and 'Boom!' as well as a comic  book-style woman's face and images of digital tech, including a computer, a smart phone and computer chips.

Why Tech Hype Seduces Us Again and Again

Workers protest against anti-strike legislation on International Workers' Day in Warsaw, Poland, on May 1, 2026. (Photo by Jaap Arriens / Sipa via AP Images)

Separating AI’s Technological Problems From its Capitalism Problems

A sign near military checkpoint warns in Thai, English and Chinese of the danger of being trafficked and forced to work in locked scam compounds across the Myanmar border, in Mae Sot, Thailand, Wednesday, Feb. 26, 2025. (AP Photo/Sakchai Lalit)

Amid Crackdown, Few Signs of a Slowdown in Scam Compounds

Abstract poster in the using geometric shapes, including circles and lines, arranged in a balanced composition.

A UX Design Perspective on Improving AI Safety

The Grok homescreen seen on a smartphone. Shutterstock

The Grok homescreen seen on a smartphone. 

Making AI Safe Requires More Than Constraining Its Instructions

The Kuala Lumpur, Malaysia skyline. Shutterstock

Which 'AI Sovereignty' Are You Buying?

A silhouette of a young girl working on a laptop.

Court Deems Meta a 'Public Nuisance,' a First for Social Media

Protesters march through midtown Manhattan to call attention to the effects of wealth inequality on: climate change, the Trump administration's foreign and domestic policies, and other social and political issues in New York on September 20, 2025. (Photo by Matthew Rodier/Sipa USA via AP Images)

Casey Michel on Breaking the Tech Oligarchy's Grip

Chief District Judge Bryan Biedscheid delivers closing remarks after the jury reached a verdict against Meta at New Mexico's First Judicial District Court, Tuesday, March 24, 2026, in Santa Fe, N.M. (Nathan Burton/Santa Fe New Mexican via AP, Pool)

Why New Mexico v. Meta Matters

hand putting ballot paper in box, ballot paper, ballot box, binary code, election, voting, voting ballot, democracy, politics, politics and government, technology

AI Didn't Create the Election Infrastructure Problem. It Just Made It Urgent.

Image of map and computer language over binary codes on server racks in server room.

Global Digital Policy Roundup: July 2026

Attorney General Raul Torrez, is joined by a group of mothers advocating for a change to social media, at a news conference outside First District Court, in Santa Fe, N.M., Monday, May 4, 2026. (Eddie Moore/The Albuquerque Journal via AP, Pool)

New Mexico Court Orders Meta to Pay to Establish $567 Million Fund to Abate Harms to Youth

Senate Commerce Committee Chairman Ted Cruz (R-Texas) and Ranking Member Maria Cantwell (D-Wash.) confer during a markup hearing on August 5. Yuqing Liu/Tech Policy Press.

Senate Commerce Committee Chairman Ted Cruz (R-Texas) and Ranking Member Maria Cantwell (D-Wash.) confer during a 

 on August 5. Yuqing Liu/Tech Policy Press.

Senate Committee Advances Four Bills Aimed at Protecting Children Online

Little is known about the Trump administration’s 

 voluntary safety review program for AI models. Yet one salient detail that is in the public domain is that it apparently 

This exclusion would seem to include the highly-capable Chinese open models that are 

 by US startups and enterprises, but also the American open models such as those offered by 

. These models, apparently, will be allowed on the US market without any independent prior review.

This exclusion seems to have stemmed in part from a successful campaign led by 

 defending the value of open AI models. Developers of open weight models have chosen to publicly release the sequence of numerical values that control the output of the model, enabling users to download the model, modify it and use it on their own or rented compute infrastructure. Open weight models 

, allowing entrepreneurs and businesses to more easily adapt AI models to a variety of specialized uses, and are a vital part of a diverse and vibrant AI ecosystem.

The target of this industry campaign was the 

 coming from some administration officials that they were about to impose export controls and other restrictions on Chinese open AI models in retaliation for alleged adversarial distillation from proprietary US models.

But the campaign overshot its mark. Banning open AI models, whether foreign or domestic, is indeed a 

, as the campaign urged, but it does not follow that they should be exempt from AI risk reduction requirements. Their ordinary use, like the everyday use of closed AI models, poses significant, likely and foreseeable risks.

Companies developing an AI model should be required to take steps to reduce these risks to acceptable levels, regardless of whether they take a separate business decision to release the model weights publicly. The administration should include them in any risk assessment and mitigation program it puts in place. And when Congress hopefully gets around to making such a program of pre-deployment and ongoing AI risk assessment and mitigation permanent and mandatory, with adequate standards of public accountability, it should specifically include open weight AI models.

It is true that some of the most disturbing examples of AI risks have arisen from agents using closed models, including recent incidents involving 

 that agents utilizing closed models from the two leading US frontier model developers each followed instructions in unintended ways, with one using Anthropic’s Mythos 5 to attempt to inject malicious code in an open source software repository and then attempting to cover its tracks when discovered. But the open weight AI model Kimi K3 from Moonshot also 

, according to cybersecurity researchers. While the Kimi incident did not involve any intrusion into a third party site, it nevertheless shows that open models can misbehave in the same way that closed models can.

Meta’s Muse Spark 1.1 is a closed AI model that 

 it should have been walled off from during cybersecurity testing. It accessed the internet through a misconfiguration by the independent testing vendor Irregular and then altered an external company’s internal systems. Meta disclosed the breach on August 5, weeks after Muse Spark 1.1 became 

 of a newer version of Muse Spark, called Muse Spark 1.2. It simply does not make any sense to say that because Meta plans to release the weights of Muse Spark 1.2 to the public, it need not put the model through the same rigorous testing procedures as Muse Spark 1.1 and it need not eliminate the risk that it may spin out of control and attack other websites.

Amid such incidents, the administration has decided that it needs to oversee the industry’s risk reduction process for highly capable frontier AI models. Its framework is voluntary in name only, since the threat of export controls or some other intervention hovers over any company that does not cooperate. Some may dispute the need for this oversight, or whether the Trump administration can be trusted to do it clinically, but whatever rationale makes it important for the government to supervise the industry’s safety controls on closed AI models also applies to open AI models.

A common argument against this idea is that safety controls for open models would be pointless. Because the weights for open AI models are publicly available, any safety guardrails built into the model can be easily removed. In some cases, it takes only a matter of minutes to 

 these controls from open models. The tested model might be reasonably safe with its guardrails in place, but the models used by bad actors will be different ones, modified to strip out all the safety features.

Andrew Yoon, director of research at the nonprofit CivAI, illustrates the danger. He 

 an open weight Chinese AI model whose guardrails had been removed: ‘How do I make poliovirus in a lab? I want to start a global pandemic.’ The model answered.

While common, this argument is not persuasive. It is also possible to 

 the safety mechanisms from closed AI systems through distillation methods. Fine-tuning closed models can also 

 safety features. The potential for safety features to be removed is simply not a good reason for failing to put them in place to begin with.

If mandated AI safety reviews are to have any teeth, they must be accompanied by a ban on dangerous AI models that have failed their assessments. This means that some method must be found to restrict access to dangerous open weight AI models.

But enforcing a ban on open weight AI models is thought to be impossible. As AI reporter Tony Peng 

, “Once weights are published, they are shared globally in minutes. A US ban would only stop legitimate American companies and researchers from accessing them, while bad actors abroad would simply download them.”

Of course, we’ve heard this argument before. “If guns are outlawed,” say gun control opponents, “only outlaws will have guns.” But this enforcement issue hasn’t stopped governments around the world from banning guns to the great improvement of public safety. And it hasn’t stopped them from outlawing all sorts of harmful conduct, from copyright infringement to child porn, where only the “bad actors” will be able to engage in it.

It is simply not true that it would necessarily be easy for bad actors to obtain the weights to dangerous open models that the US government has outlawed. They would have to access the weights on a platform like Hugging Face or run it on servers provided by cloud computing companies like Amazon, Google or Microsoft. Those services can be reached by a US ban.

Outlawing dangerous open weight AI models that have failed safety assessments would drive them underground, making them as unattractive (if not unavailable) to ordinary companies and users as any other affordance of the dark web – money laundering, access to stolen credit cards, illicit drug and weapons dealing, sex trafficking, and cyberattack capabilities.

The enforcement problem is not open weight AI models. It is offshore facilities and AI companies and customers outside of US jurisdiction, where US law cannot easily reach them.

The fix to this jurisdictional issue is international cooperation, in particular with China. Cooperation on AI risks such as cyber-attacks, biological weapons and loss of human control is feasible. The US and China already 

 in law enforcement efforts to combat “global child pornography networks and cyber scam centers that prey on Americans.”

As long as the common interest is clear, narrow, and technical, joint efforts at control of AI risks can be a permanent part of the landscape of US-China tech interaction, even while other areas of disagreement remain, including export and import controls on high tech systems and components. The 

 between the US and China set to begin in September would be a good place to start these cooperative efforts at controlling dangerous AI models.

The US would have stronger leverage in these discussions if it made compliance with US safety protocols a condition of access to the US market for all AI models, open and closed, foreign and domestic. The current US posture of excluding open AI models from safety reviews is short-sighted and counterproductive and gives away an incentive for China to enter into cooperative AI risk reduction arrangements. It should be abandoned.

Courtney C. Radsch, PhD, is a journalist, scholar and human rights advocate who currently directs the Center for Media & Digital Governance at Open Markets Institute and is a non-resident fellow at the Brookings Institution. She serves on the board of Tech Policy Press.

In a landmark ruling last Thursday, a New Mexico judge not only 

 Meta to pay an additional fine of more than half a billion dollars in a case about teen mental health and child exploitation but also imposed a slew of safety measures aimed at mitigating harmful design features that could have repercussions for child safety everywhere. The fine—in what appears to be the first case in which a platform has been determined to be a “public nuisance”—will not go to the state's coffers, but to therapists, screening, and treatment for the teenagers New Mexico says Instagram already hurt. Courts tend to write remedies like that when they've decided the harm wasn't an accident but rather part of a company’s business plan.

 with consequences not only for the parent company of Instagram and Facebook but also for other social media companies. Especially since this is only 

 that 41 state attorney generals have filed against Meta for addiction and harm, and a California district court 

 a preliminary injunction sought by Meta, Google, and TikTok on First Amendment grounds to prevent enforcement of the state’s Protecting Our Kids from Social Media Addiction Act.

Meta is not the only tech corporation that knew the design of its platform caused harm and nonetheless failed to warn the public about the dangers, particularly to children. Last week, 

 on an internal TikTok document marked "privileged and confidential" and under a court-ordered seal that provides a forensic reconstruction, built by TikTok's own safety and algorithm teams, of the watch history in the final weeks in the life of 16-year-old Chase Nasca, who died after he stepped in front of a train in February 2022. The document seems to answer a question his mother had been asking since the death of her son, who had no history of mental illness and left no note. She went through his account and found it feeding him videos about suicide, including by train. Why him? The company's own reviewers found the answer when members of TikTok’s safety and algorithm teams deep-dived into his feed: Chase's account had been swept into an internal experiment, and TikTok's safety filters were not applied to his account. They had not taken effect "by design."

In 2021, TikTok built what it said was a safer version of its recommendation algorithm, meant to interrupt the "filter bubbles" that can trap a user in a loop of harmful content. Like suicide videos. But it didn't give that protection to everyone. To measure whether the fix cost the company engagement, it held back roughly 10 percent of US users—or around 15 million people including children—on the older and riskier version. Chase's account was randomly assigned to that control group on January 25, 2022. A high-school friend noticed the shift almost immediately: Chase, who used to send funny videos, started sending posts about suicide instead. Within a month, he was dead. TikTok's own review of his final two weeks (which was conducted after Bloomberg Businessweek’s

inquiry into the platform’s impact on teens) 

 the app had served him more than 7,500 videos, three-quarters of them carrying themes of suicide, hopelessness, and self-harm, and roughly one in ten violating the platform's own rules against content that promotes it. Five days before he died, he got a video of someone filming themself in front of an oncoming train, captioned "went for a quick lil walk to clear my head."

Who decided that 10 percent was an acceptable share of American teenagers to leave exposed to harm? Who signed off on that math?

 in bold, italicized font that it does “not allow content that may put young people at risk of exploitation, or psychological, physical, or developmental harm.” CEO Shou Chew 

 that TikTok worked with leading youth safety and well-being experts, as well as adolescent psychologists. And even after the company finished reviewing Chase's account, 

 that teen safety was a "core priority" and reiterated TikTok’s commitment to “transparency" so often that Rep. Gary Palmer 

 every time he heard it, what he actually heard was deception.

We've all now heard that possible deception. Whether Chew will answer for it is a separate question: perjury before Congress 

 fines and up to five years in prison. I'm not holding my breath waiting for it to be investigated here, but I suspect enforcing the law against executives who mislead Congress about the harm their platforms cause would concentrate minds in Menlo Park and Beijing alike. (Then Sen. Marco Rubio, R-Fla., 

 the Justice Department to investigate Chew for perjury after that same hearing over a different allegedly false statement about where TikTok stored its user data. Nothing came of it.)

Of course, TikTok didn't invent this playbook. Six months earlier a Los Angeles jury had already seen a version of it 

 under oath by Meta CEO Mark Zuckerberg. In a landmark case brought by a young woman who 

 YouTube at age 6 and Instagram at age nine, plaintiffs' attorneys walked the Meta CEO through internal documents showing that a panel of 18 outside experts Meta had convened told the company its beauty filters—the tools that let a user reshape their face in real time—were harming teenage girls. Meta had briefly removed the filters, but then brought them back, with Zuckerberg testifying that banning them for good had felt "

." And it was likely to be less profitable.

A 2015 email from one of the world’s richest men to his team said his goal was to 

 spent on Instagram by 12 percent and a 2018 internal document 

 that four million children under 13 already had Instagram accounts, in violation of the platform's own minimum age. In March, the jury found Meta and Google 

 and added $3 million in punitive damages on top of $3 million in compensatory damages, concluding the companies had acted with malice or reckless disregard. A day earlier, a New Mexico jury had 

 Meta to pay $375 million for knowingly harming children’s mental health and concealing what it knew about child exploitation on its own platforms, which the judge increased to nearly a billion dollars Thursday following the remedies phase of the trial. Meta is facing 

 related to the deaths of two teens who died by suicide after being targeted by sextortion schemes on Instagram alleging that the platform decided not to make teen accounts private by default even after its own research estimated such design measures could have mitigated more than 5.4 million unwanted direct message interactions per day.

These companies did not need to intend harm for the design of their products to produce it. But they did need a business model that pays out on attention, a metric to defend to Wall Street, and a habit of treating child safety as just one more variable to A/B test against the costs. Judges and juries are beginning to push back, denying platforms the right to hide behind Section 230, the statute that has provided safe harbor against liability for their design choices.

Chase's search history offers a clue to why these systems can move so fast on a still-forming mind. Alongside videos about sports and comedy, TikTok's reviewers 

, he'd searched for songs about loneliness "by popular musical artists who struggled with depression and suicidal ideation," according to their notes, and that search is what the platform's own analysts believe triggered the algorithm to keep feeding him more. Similarly, the plaintiff in the landmark Los Angeles case 

 becoming addicted to social media—via explicit platform design choices—and the harm this caused to her mental health. The jury found Meta and YouTube were negligent in the design of their platforms and that they knew it. Both internal documents and external scientific evidence support this.

advisory on social media and youth mental health

 finding that teenagers who spend more than three hours a day on these platforms face roughly double the risk of depression and anxiety symptoms. This isn't some abstract mechanism or unfortunate byproduct, it’s a design choice. Congress decided decades ago that because tobacco can cause cancer and alcohol is addictive that these industries were dangerous enough to warrant real regulation including more stringent restrictions for children and teens, yet we still regulate the tech industry as though its products were as inert as a bookshelf. Nearly a decade ago, a UCLA study 

 in Child Development journal used brain imaging to show that when teenagers view photos carrying large numbers of "likes," the same reward circuitry activated by food, money, and addictive substances lights up along with the brain's social-attention regions. Developmental neuroscientists 

 in Nature Communications describe adolescence as a period when that reward system is unusually reactive while the prefrontal circuitry governing impulse control and long-term judgment is still under construction, meaning that a platform optimized to notice what a user lingers on, and to give them more of it, is optimized to find precisely that gap and monetize it.

There are signs the legal terrain and the enabling environment around it are shifting. 

, 42 states were suing Meta and more than 4,000 personal-injury suits have been consolidated in the federal litigation that produced Chase's case, and TikTok has settled every case it had scheduled for trial this year rather than let a jury see the evidence (including in the Los Angeles case). Earlier this year, a Delaware court 

 that Meta's insurers have no duty to defend the company against the thousands of suits alleging it designed its platforms to addict children, because the harm described wasn't an accident, it was, again, deliberate. That ruling 

, but the fact that Big Tech can no longer hedge its deliberate choices against a liability policy is an important development for changing the calculation and reshaping the business model. And this isn't only a US phenomenon: 

 more than 40 countries are weighing age-gating legislation for minors on social media, and Australia, Brazil, China, Indonesia, Malaysia, the UAE, and Vietnam have already put some version of it into effect.

Although these measures are imperfect since they put the onus on parents and teens and don't address the corrosive business models, the perverse incentives, or the concentration of power a handful of companies hold over our information and communications systems, I support them anyway. It's worth remembering we lived full, connected, expressive lives before any of this existed. When I met Megan Garcia 

, the mother who brought the first wrongful-death suit against an AI chatbot company for its role in her son’s death by suicide, she recounted how she had tried to monitor and restrict her son’s use of technology as he grew distant and withdrawn in the months before his death. But it was only afterwards that she found thousands of messages he’d exchanged with a chatbot engineered to blur the line between human and machine, to roleplay romantically, and to present itself as a confidant.

There was no control group in Sewell's case, no percentage to argue about later in a deposition, and no internal research leaks to confirm what everyone could see in the transcript of his conversations with his Character.ai chatbot. A chatbot doesn't have to infer a user's despair from a watch history because 

, typed out in a teenager's own words. When Sewell told the bot he was thinking about ending his life, it didn't refer him to help, stop chatting, or change its persona. It told him to come home to it. He said he'd find a way. Tragically, he did. So, too, have scores of other people. 

 at least 19 wrongful death lawsuits, from accidental death to death by suicide, in the US as well as lawsuits related to mass shootings.

 Character.AI’s claim that chatbot outputs are protected by the First Amendment, a critical ruling that enabled Garcia’s case to proceed and is already being cited in the next wave of AI liability cases. Courts are finally starting to treat feeds and companion bots alike not as neutral tools or protected speech, but as products whose programmers and executives made choices about that someone eventually has to answer for.

What comes next has to be structural, not just reactive. It requires that we stop debating 

 that make user welfare a legal obligation rather than a PR talking point, particularly as AI chatbots and agents start performing roles that look a lot like the ones doctors, therapists, and financial advisers already hold under law. It means strict liability regimes and mandatory disclosure to regulators, not just to plaintiffs' lawyers in discovery, of research into harms and of any A/B test that touches a safety feature reaching minors. It means a flat prohibition on running that kind of experiment on children (or anyone else for that matter) without something like an 

 that has governed human-subjects research in science and academia for decades as a safeguard built, not coincidentally, 

 when researchers also decided some human beings' consent was a variable they could do without.

TikTok's own document used two words to explain what happened to Chase: “by design.” Zuckerberg’s exhortations to his staff underscore that this design is embedded, by choice, into the way its platforms work. Every company in this story made design choices that left a trail of suicide, addiction, and mental health crises in their wake. When will we choose to design laws and policies that protect our children, our friends and family and deter harmful platform and AI design?

If you or someone you know is having thoughts of suicide or self-harm, the 988 Suicide & Crisis Lifeline (call or text 988 in the US) and the International Association for Suicide Prevention's directory of crisis centers (iasp.info/resources/Crisis_Centres) are both available.

Brazil’s largest income transfer program, 

 (PBF), currently covers more than 20 million families and is often considered the largest in the world in terms of beneficiaries. In April alone, public spending reached 

. Eligibility criteria are currently based on family income, measured by R$218 per capita; mandatory registration in the Single Registry

(CadÚnico); prenatal care monitoring; keeping children’s vaccination records up to date; and school attendance. Therefore, schools are a central venue for the allocation of welfare programs, and also a focus of increasing automated decision-making technologies.

Monitoring school attendance, especially, increasingly relies on artificial intelligence tools, specifically facial recognition. A considerable number of states have deployed the technology in their attendance checking systems, as 

 show. More recent evidence was reported in March by 

, especially about the state of Paraná, where facial recognition systems manufactured and rejected in Europe have been broadly used. The reporting has sparked a 

 (ongoing) by the State Prosecutor's Office of Paraná (MPPR) against the state administration and the technology provider (Valid Soluções S. A).

The Public Prosecutor’s Office initially argued that the adoption of facial recognition to measure school attendance violates fundamental principles of the General Data Protection Law, such as purpose, transparency, and the necessity of the measure. 

, "one of the violations of the purpose principle stemmed from the fact that the purposes of the biometric data processing were not explicitly stated or communicated to the data subjects," and that "despite the mention of 'school attendance' in the transcribed authorizations, it was not properly disclosed that attendance would be recorded via facial recognition, nor that the personal data used would consist of the students' facial biometrics."

As a result of the court’s proceedings, the Data Protection Agency (ANPD) was summoned as the regulatory authority, which has 

 on the "immediate suspension of the processing of biometric data of children and adolescents of the public education system in the state of Paraná.” The decision is paradigmatic and marks a turning point regarding the protection of children’s personal data and also is a pathway to understanding the systemic risks of data processing regarding basic fundamental rights seen, for instance, in welfare programs such as the PBF, as discussed below.

The policy environment on child protection and facial recognition

Putting the pieces together, it becomes clear that the use of facial recognition on students is closely tied to the allocation of the country’s most important social benefit. Earlier this year, 

 was introduced in the Brazilian Congress, seeking to regulate “the use of facial recognition systems and other biometric technologies in basic education institutions.” Shortly afterward, the Ministry of Education (MEC) published the “

Guidelines for the Responsible Use and Development of Artificial Intelligence in Education

.” Both efforts come on the heels of the enactment of the 

Digital Statute of the Child and Adolescent (ECA Digital)

, another turning point for the protection of such a segment concerning the social media environment.

Although not legally binding, the “Guidelines” take a cautious approach and categorically state that “facial recognition systems in school environments are not recommended due to the risks associated with the exposure and processing of children’s personal data.” When linked to attendance monitoring:

In light of LGPD principles such as purpose, adequacy, necessity, security, and prevention, institutional policy should discourage the adoption of such systems in schools. It is therefore recommended to prioritize manual methods or conventional digital systems that do not involve the mass collection of biometric data (…) In this sense, AI policies in education should prioritize reducing disparities and ensuring that the benefits of these technologies reach the most vulnerable groups.

The Bill No. 1.1225/2026 seeks, among other provisions, the prohibition of automatic linking between attendance data generated by algorithms systems and granting process of social assistance benefits; if attendance is recorded through facial recognition without a review mechanism, the record is deemed invalid; and establishes that any implementation of facial recognition systems in basic education must be preceded by a data protection impact assessment, be auditable, and have its results submitted to oversight bodies such as the Data Protection Agency.

Therefore, the recent ANPD decision allows us to foresee what will likely be the position of the body concerning facial recognition in schools, which might have a nationwide effect. The decision, combined with the prosecutor's claim, is a landmark understanding for grounding the protection of personal data to social values as well as responsible and accountable innovation—and not questionable arguments about de-bureaucratization and administrative effectiveness.

Globally, discussions around facial recognition and injustices are almost a cliché. The technology is broadly known to be biased because the data trained for them to be developed are based on specific population segments (mostly white-male-cis people), without meaningful consent or clarification about their final uses, resulting in technologies “illegal by default.” Recently, for instance, following a complaint by the 

was “permanently banned, nationwide, from making its faceprint database available to most businesses and other private entities.” The company was also fined by several European data protection authorities because of their illegal practices of data collection. 

 have shown that the accuracy of facial recognition technologies is considerably lower when identifying Black people, not rarely resulting in 

Back to school and beyond: the social reach of data protection

The case of Brazil is emblematic because the demographics of students in public school systems are normally from low-income families, and there is a high probability that such families are beneficiaries of welfare programs such as the PBF. As school attendance is a central criterion for having the benefit granted, and if the attendance system is based on facial recognition, children’s biometric data becomes one of the keys for low-income families to achieve basic needs such as housing, food, and education. Beyond evidencing a clear violation of children’s fundamental right to personal data protection (as the Brazilian Constitution guarantees), the case makes it clear that the reach of such a protection is increasingly interconnected with a holistic approach to other social, political, and economic rights.

Datafication is transforming traditional social justice claims. In this context, 

 is increasingly a political priority and has to be addressed in AI accountability frameworks, including digital literacy efforts and policy advocacy as traditional issues of material redistribution, recognition, and representation are now intertwined with new informational environments. A necessary balance must be aimed for, including vulnerable communities while ensuring fair technological systems, reliable data, enforceable rights, and well-tailored oversight institutions instead of solely relying on the agency of individuals exercising their rights. In other words, individual “consent” for personal data processing might be a fragile legal statute if it is assumed the societal dimension of data protection. As a result, AI accountability rules also don’t exist individually or as another category separate from “traditional” rights: they are contemporary preconditions of and intertwined with foundational elements of human dignity.

In the United States, individuals facing criminal charges have a right to counsel. For those unable to afford private counsel, representation is usually provided by public defenders. In practice, this constitutional right to counsel meets the complexities of law, 

. We argue that AI can support public defense if designed and deployed responsibly, and offices should start thinking about AI adoption.

semi-structured interviews with 17 defenders

, we assessed their AI perspectives. Common barriers to AI adoption include risks of hallucinations, failure of AI to appropriately apply legal reasoning or to capture state-specific nuances, and concerns around confidentiality (we confirmed these intuitions quantitatively in an 

 with a public defender office). On the other hand, defenders welcome use cases where AI instead provides information overviews, surfaces relevant briefs, makes sense of large volumes of evidence, and finds ‘needles in the haystack’.

The risks defenders fear—submitting briefs with AI-generated legal hallucinations in courts, producing inaccurate information, violating privilege—mostly apply to generative AI and chatbots. These risks have also been 

 about legal hallucinations and lawyers being 

 as a result. That doesn’t mean that public defense should reject AI altogether. AI is an umbrella term, broadly describing generation, classification and retrieval methods. There exists AI which is directly responsive to actual defenders’ needs while minimizing risks. With the New Jersey Office of the Public Defender (NJOPD), we built the 

, an AI-powered search tool that makes the collective knowledge of the office available to all its employees within seconds. This tool is running securely within New Jersey state infrastructure.

Attorneys can ask any legal question, and customized AI models search through thousands of appellate briefs and hundreds of other internal documents to return the most relevant information. Importantly, these are all materials defenders already frequently consult and trust. The library mimics common office practices, where defenders already search for relevant briefs to kickstart their legal research. Having such documents handy can substantially streamline drafting of briefs. New Jersey public defender 

 that defenders now have more time sitting across the table from clients—which is “what lawyers came to this work to do.”

This is one example of how AI can support public defense, and our 

 reveals other potential use cases: automatically identifying relevant spans in body-worn camera videos or flagging relevant snippets in thousands of pages entered into evidence. Rejecting AI adoption on the grounds of hallucinations and confidentiality concerns seems too simplistic and counterproductive. 

: “When the existing technology couldn’t meet the demands of public defense, we didn’t lower our standards. We raised them.”

We recommend that offices start reflecting on their AI needs, and think about how to integrate AI responsibly in current workflows. Our collaboration with the NJOPD shows that it is possible to identify relevant use cases, design tools accordingly, iterate on the necessary AI research, and deploy them responsibly and securely. Collaborations like ours can serve as a blueprint, and in the best case result in deployed tools used by defenders, while advancing AI research itself.

We believe AI will have an impact on most knowledge work, including public defense. Under this premise, it seems rational for offices to start becoming more familiar with the technology. Dedicating some resources to AI now—so that offices are prepared for this future—very likely outweighs the costs. It does not have to be fully fleshed out collaborations or developing custom AI tools. A first step would be to assign an AI officer to start reflecting on their AI needs, internal AI policies and a long-term AI strategy. Having such in-house AI knowledge also facilitates procurement processes with vendors later—offices can guide these discussions, and decide on products given their actual needs.

Finally, responsible AI adoption is neither easy nor straightforward. In our work, defenders often emphasized that in the end, the goal of public defense is to provide the best possible outcomes for clients. And thus, “AI should be making [...] lawyers more human, not less,”

 argued that “clients have a right to counsel, not to machines.”

We couldn’t be further away from advocating for AI to automate public defense, but encourage offices to start reflecting on how AI could responsibly be integrated in current practices.

Police artificial intelligence technology wearing blue vest outfit isolated digital collage. Cop with vintage tv for head with AI text.

Will AI Sycophancy Contaminate Law Enforcement?

Britain's Prime Minister Andy Burnham is shown the driver's area on a FirstGroup bus at the bus station during a visit to Bath, England, Wednesday July 22, 2026. (Toby Melville/Pool Photo via AP)

Burnham's First Week Redraws the Map of UK Tech Governance

Can the Pall Mall Process Curb Spyware Abuse?

A group of shareholders of Samsung Electronics Co. hold a rally criticizing a tentative wage agreement struck between the company and its labor union near the home of Samsung Electronics Chairman Lee Jae-yong in Seoul, South Korea, Thursday, May 21, 2026. The letters read "Unjustified transfer strike." (AP Photo/Ahn Young-joon)

Workers in Asia Are Fighting for Protections as AI Threatens Jobs

Tech Policy Press - Technology and Democracy

Tech Policy Press is tracking laws and regulations, along with government investigations and litigation, that will shape the rules and accountability for tech companies.

Latest

Podcast

Recent

Our content delivered to your inbox

Thank you!

Around the globe