Who Participates in Politics on Social Media, and Why?

Europe Wants to Regulate AI Friends. But How Do You Curb Dependency?

Taoiseach Micheal Martin and European Commission President Ursula von der speaking during a press conference in the Main Quadrangle at University College Cork, as Ireland begins its presidency of the EU Council. Photo: Liam McBurney/PA Wire IRISH (Press Association via AP Images)

At Ireland’s AI Summit, Tech Industry Dominates as Civil Society Sits Out

A vintage court room with many chairs, benches, and a neon lit ceiling, and black and white portraits on wood panelled walls. 

A courtroom of the US District Court for the Northern District of Illinois. 

Nippon Life v. OpenAI and the Future of AI Legal Work

A grafiti drawing of a white skeleton on an asphalt street with a sewer grid placed direclty where the ribs would be.

 is shared by diegoluis on Flickr (CC-BY)

Tech Hype is Cannibalizing the Future

The street front of a mobile money agent with an Airtel umbrella and sign in the foreground. One person is sitting to the left, reading, another person can be seen from behind inside the store. 

 in Gulu, northern Uganda in 2016 shared by Fiona Graham/WorldRemit on Flickr (CC BY-SA)

In Africa, Digital Payments Could Unlock Tax Revenue if Data is Protected

A pile of narrowly shredded white newspaper with black writing that is not legible.

Cutting Local Journalists, McClatchy Embraces AI Content

Disinformation Conference in Europe Caught Up in Transatlantic Politics

President Donald Trump gestures to Elon Musk as he speaks to reporters at the White House in Washington, Tuesday Sept. 29, 2026, after meeting with top executives of AI firms. (AP Photo/Jacquelyn Martin)

It's Time to Regulate the AI Development Process

Britain's Prime Minister Andy Burnham gestures as he speaks next to British AI Minister Kanishka Narayan during a business and investor roundtable before attending the 81st United Nations General Assembly in New York City, Tuesday Sept. 22, 2026. (Toby Melville/Pool Photo via AP)

UK Wants to Lead the World on AI Safety. It Can’t Decide Where To Start

Library shelves with books facing each other, with light against a wall in the distance. It says "NONFICTION" on each shelf. 

 shared by Orin Zebest on Flickr in 2007 (CC-BY)

Can AI Help Fix Peer Review Instead of Breaking It?

A white banner with dozens of names and ages written in black is held by a group of teenagers in front of a building, seen between the dark red leaves of a tree. 

A banner lists the names of young people advocates say died because of social media outside the Ronald V. Dellums Federal Building and U.S. Courthouse on Aug. 18. A settlement in a multistate lawsuit accusing Meta of harming children was reached there on Aug. 26. (APPhoto/NoahBerger)

No, the Meta Settlement Isn’t a First Amendment Problem

Professsor Yoshua Bengio, co-chair of the UN International Independent Panel on AI attends Security Council meeting on Artificial intelligence and international security at UN Headquarters in New York, NY on September 23, 2026. (Photo by Lev Radin/Sipa via AP Images)

The UN's AI Panel Sees Misalignment. We See Corporate (Mis)Behavior.

Meta CEO Mark Zuckerberg listens as President Donald Trump speaks in the East Room of the White House in Washington, Tuesday Sept. 29, 2026, during a meeting with top executives of AI firms. (AP Photo/Jacquelyn Martin)

Meta's Multibillion-Dollar Settlement Left a Lot to Be Desired

March 10, 2026: US Customs and Border Protection officer walks through a busy airport terminal concourse with travelers in Miami, Florida. Shutterstock

March 10, 2026: US Customs and Border Protection officer walks through a busy airport terminal concourse with travelers in Miami, Florida. 

Circuit Court Ruling on Border Searches Deals Serious Blow to Privacy Rights

Metal full body height turnstile gates with a red X in the middle, above a narrow black panel. In the blurred background, between the bars, refugee housing containers, a mountain and the sunset can be seen in the distance. 

Biometric fingerprints and police screening are required upon exit and entry to the refugee camp on Samos Island in Greece. Photo from September 21, 2021 (Nicolas Economou/NurPhoto)

Human Rights Should Still Matter When Countries Buy Tech That Kills

Illustrative photo captured during the plenary session of October 2025 in Strasbourg, France.

“Behind the scenes - European Parliament in Strasbourg” by European Parliament, 

The EU’s Tech Regulation Paradox is Getting Harder to Ignore

President Donald Trump holds a press conference on the North Lawn of the White House with Speaker of the House Mike Johnson, Mark Zuckerberg, Elon Musk, and other tech leaders on September 29, 2026. (Photo by Andrew Leyden/NurPhoto via AP)

An AI Accord, a New Czar and an Autonomous Warfare Command

A "No Data Center" yard sign is seen in Allenwood, Pennsylvania on July 7, 2026. The sign is in opposition to a proposed 300-megawatt data center at nearby Great Stream Commons industrial park in Gregg Township. (Photo by Paul Weaver/Sipa via AP Images)

Pennsylvania is a Swing-State Case Study for Data Center Politics

Meta CEO Mark Zuckerberg listens as President Donald Trump speaks to reporters at the White House in Washington, Tuesday Sept. 29, 2026, after meeting with top executives of AI firms. (AP Photo/Jacquelyn Martin)

Why Social Media Platforms Should Adopt a 'Militant Democracy' Framework

A central question of the last decade is what the emergence of social media as a primary locus of communication has meant for political discourse across the world. What social movements owe their existence and power to social media? What are the affordances and mechanisms by which they are able to spread? Is the public becoming more informed through social media, or led astray by misinformation? And whose voices are and are not heard on these platforms?

A recently published paper explores one discrete but important piece of this puzzle: the relationship between encountering political content on social media, being better informed on current affairs, and participating in political discourse on social media.

A new window to the world? Associations between political social media content, political knowledge and participation on social media

,” Cato Waeterloos of Hasselt University in Belgium conducted an empirical study to interrogate a posited theoretical framework. In this model, users who are exposed to political posts on social media show increased participation in this online discourse via the “mediating mechanism” of having increased their knowledge about the issues.

Chart showing hypotheses from this study. H1 suggests that exposure to political content on social media will increase social media political participation (SMPP). H2 suggests exposure to political content will increase current affairs knowledge. H3 suggests increased current affairs knowledge will encourage SMPP.

Waeterloos fine-tuned her approach to studying this set of relationships by zeroing in on the kinds of political content common on social media, the salient knowledge that this is likely to improve, and participatory activities common on social media. She went beyond hard news about institutions and politicians in defining political 

, including “lifestyle-oriented political topics” like climate activism and gender equality. This broadened the concepts included, but 

about those concepts was also narrowed to “knowledge about recent political and societal events” in keeping with the way that social media prioritizes live issues over knowledge of history and longstanding systems. Finally, social media political 

was taken to include both original posting and “low-threshold actions,” like sharing and indicating event attendance, which can also be significant when viewed through the author’s political socialization lens.

Waeterloos surveyed 2,520 individuals in Flanders, Belgium, in December 2023. The subjects were asked to estimate how many times they saw information on social media about six specific political and societal topics which had been identified as commonly present on the platforms in the previous three months. They were then quizzed with eight multiple-choice questions about local and international political stories. Finally, social media political participation (SMPP) was assessed using a measure 

 by the author and collaborators, where the subjects were asked how often they had participated in each of five engagement activities (posting, sharing, etc.) over the last year. The study also measured control variables including political interest and socioeconomic status.

The findings supported the author’s hypothesis that exposure to political content would be positively associated with both SMPP and higher levels of current affairs knowledge (H1 and H2, as labeled on the diagram above). This suggests that, when the metrics are dialed in for the type of political discourse most often occurring on social media, people may well be learning (true things!) about current events from the content on the platforms.

However, the results did not support Waeterloos’s other hypotheses: higher current events knowledge was, in fact, negatively associated with SMPP. This is inconsistent with the broader premise that learning about politics on social media leads to increased active political engagement online (H3) because it increases knowledge about the issues (H4). Rather, it indicates that “people who objectively know more about current affairs are actually less likely to express themselves politically on social media” and therefore that those less well-informed are more likely to contribute to the discourse.

The author suggests three possible explanations for this mismatch. All three are rooted in academic literature and have some intuitive appeal for those attuned to the nature — and problems — of online discourse.

Waeterloos first suggests that some people, despite being informed about the issues, are still reticent to speak out, perhaps because they are fearful of the enduring nature of their contribution and the online culture of “discursive accountability practices.” Curiously, she links this to the subjects’ possible awareness of their lack of background context on political topics, as the study only assessed current affairs knowledge. While this is one possibility, these fears may reasonably persist even for those well-versed in the history and likely implications of events; online retribution can reach any user who says something unpopular with any faction, even if their expertise is unimpeachable.

Related to this is the second proposed explanation: that SMPP is not associated with objective knowledge about political issues, but rather with subjective knowledge. Overconfident voices may in fact dominate on social media, and this might deter some objectively knowledgeable users from engaging in shared discourse. If this is true, Waeterloos tellingly observes, it would cast doubt on “certain normative assumptions in democratic theory, which classically assume informed and knowledgeable citizens to be the ones participating in politics.”

” are common online: small groups with deep knowledge and interest in a particular topic. If enough of the social media discussion of political issues occurs in these settings, then, even if the theoretical model were basically sound, broad current affairs knowledge would not be the correct factor to measure.

It is important to note that this is not a causal study, and so the findings could only align or not align with the proposed model, not prove or disprove it. Although the survey was constructed based on data, and the exposure topics matched the knowledge questions, there may still be problems with the reliability of responses (did they really see the content?) and the presence of the specific facts in the content that was encountered. There may also be significant differences between people who seek out political content on social media and those who encounter it incidentally.

In addition to these factors that were mentioned in the paper, empirical academic studies of this type on social media are often limited in other ways. The research did ask subjects which platforms they use to ensure that this was not a confounding factor. However, the findings were published two and a half years after the data was collected. In that time, as 

 have noted, the participatory nature of social media has seen a significant transformation, along with some of its key affordances.

Social media is also dependent on the underlying culture of the users, a fact underscored by a 2024 survey showing that views diverged widely between countries on the question of 

if social media is good or bad for democracy

. The subjects of this study all lived in one linguistically distinctive area of a single small country, and it is not clear that the culture of political discussion on social media would be the same in other places.

 quarters, this study gave a ray of hope that people do indeed learn true facts about current events from content on social media. However, it also showed that a straightforward model where citizens first become politically educated and then step up to become knowledgeable political actors on social media may be flawed. Sadly, what flows from the “window to the world” may not be the most well-founded contributions.

Waeterloos gives many worthwhile suggestions for further study to better understand how and why people participate in political activity on social media. While our knowledge in this arena no doubt remains incomplete, those involved in product and policy work within and outside of platforms may already look to this work to inspire proposals to improve democratic mechanisms online. What are the cultures, affordances, or elements of algorithmic design that might underlie the apparent negative relationship between knowledge and participation, and how can we change these to incentivize and amplify the most well-informed voices for the benefit of all?

Liz Carolan is a Tech Policy Press fellow.

As Europe enters the final stages of negotiating the 

, Ireland is preparing to welcome the world’s top technology executives to an exclusive AI summit in Dublin next week. Ireland is currently chairing the final stages of negotiations over the package of measures designed to “simplify” Europe’s rulebook.

 will take place at the Royal Dublin Society, a cavernous, historic venue a few hundred meters from the US Embassy. It will be attended by at least three European Commissioners, four Irish Ministers, and an undisclosed number of European officials. Prominent speaking slots at the event will feature US AI firms Google, Microsoft and OpenAI, with Meta, IBM, Intel, Nvidia, Amazon, Anthropic and AMD also represented on panels, alongside major European-founded firms like Stripe, SAP and Mistral.

The summit has been positioned as a centerpiece of I

reland’s Presidency of the Council of the European Union

, the six-month rotating chairmanship of the body that represents member-state governments. It comes as the European Union is modifying its regulations to increase global competitiveness, a process it calls 

strengthen economic and digital sovereignty 

by supporting the growth of European rivals to US and Chinese industries.

These priorities are converging in the final stages of the negotiations over Digital Omnibus, which Ireland is chairing as holder of the Council Presidency. This 

 would amend parts of Europe’s digital rulebook, including the GDPR. Critics have 

European Commission Executive Vice President Henna Virkkunen is set to open the summit, alongside Irish Taoiseach (Prime Minister) Micheál Martin. Virkkunen will arrive in Dublin having set out a position on AI regulation at odds with the US.

 the Commission would keep pursuing an international agreement on AI safety and security despite US resistance, and the EU has backed 

 for an international body to monitor AI safety. Meanwhile, the White House has pushed back against AI regulation and announced 

a voluntary safety accord with six US AI and chip companies

, five of which will be represented on stage in Dublin.

 that industry speakers dominate the summit’s agenda. A panel called “AI Transforming Health: From Discovery to Patient Care” does not include health authorities or patient representatives, instead featuring three pharmaceutical companies, Johnson & Johnson, Novartis and Sanofi, alongside Nuritas, an Irish firm that uses AI to discover peptides.

Tech Policy Press analysis found that 82% of speaking slots on the official agenda, 37 of 45, are executives representing companies or business groups. The remaining eight slots go to three European Commissioners, the Taoiseach and three Irish Government ministers, and a single trade union representative. No civil society organizations appear on the agenda, and no academics are listed among its speakers.

The trade unionist was a late addition. In July, 

 the lack of worker representation at the summit, including on its planned panel on AI in the workplace. Summit organizers subsequently added a representative of the Communications Workers' Union (CWU) to the program, where they will share a 10-minute slot, "The Changing World of Work", with employers' group Technology Ireland.

The tone of the summit contrasts with broader debates happening around AI, some of which will be held at independently organized events happening in the days around the event. The 

 is hosting a “complementary public-interest conversation” alongside the summit that will focus “not simply on what AI can do or how quickly it can be deployed, but on the human values, institutions and ways of life that technological progress should ultimately serve.”

,” aims to be “far more representative of the real experiences of AI on workers and communities than the government's flagship event,” organizer John Bohan of the CWU Union told Tech Policy Press. That event is explicitly critical of the “mainstream narratives on AI,” with Bohan saying debates about AI are “massively one-sided in favor of corporate interests.” That two-day event will feature groups including 

The Counter Summit organizers also criticized the Irish government’s approach to tech firms, telling Tech Policy Press that its “industrial policy approach” has been to “pander to the needs of the tech giants.” Critics attribute that pandering to pressure to retain US tech firms that dominate Irish economic and fiscal policy. Just two tech firms — Microsoft and Apple — 

made up about 40% of Ireland’s total corporate tax

 receipts last year. This allowed the government to announce a “giveaway” 2027 budget this week with income tax cuts, social welfare increases, and subsidies on home energy costs. Yet the official budgetary watchdog, the Irish Fiscal Advisory Council, 

 in response that these giveaways entrench dependency on “high-risk corporation tax”, amid fear that Ireland will become less attractive to US firms as valuable tax schemes run out by 2030.

 are expected to attend the invitation-only summit next Wednesday, and the guest list has not been made public. Several events on the margins suggest officials from across the bloc will attend, including the 

, which brings together member states' digital ambassadors and is 

holding an informal meeting in Dublin the day after

It is believed that the invite list is managed by the hosts, Ireland’s economic development ministry, together with the Department of Foreign Affairs (DFA). The DFA is coordinating the presidency, a process involving complex 

, with press accreditation centrally managed for the six-month period. Tech Policy Press has heard from a number of civil society and media groups who have struggled to get on the list.

The Irish Government has spent months preparing for the summit, which it sees as an opportunity to bring together the tech industry its economy relies on and that Europe hopes to foster and grow, with the European policymakers it plays a central role in stewarding. But while officials have kept tight control of the agenda and guest list, events outside the venue may intrude on the day.

 that coincides with the summit, demanding pay increases funded from the surplus tax revenues generated by Ireland's foreign direct investment. On a day designed to showcase Ireland's relationship with the tech industry, workers seeking a share of its proceeds may prove harder to keep off the agenda.

A courtroom at the US District Court for the Northern District of Illinois. 


Artificial intelligence (AI) writes a rapidly growing share of pro se litigant filings, with two 

 finding AI-flagged writing in over 18 percent of such filings. That share that has risen each year since ChatGPT’s release. A case pending in the US District Court for the Northern District of Illinois, 

Nippon Life Insurance Co. of America v. OpenAI Foundation

, is among the first cases to ask whether AI-assisted legal briefs constitute the unauthorized practice of law (UPL) and whether the AI company that made it is liable.

Illinois has two statutes regarding UPL: the 

, which dates to the mid-1800s and bars unlicensed persons from practicing law, and the 

Corporation Practice of Law Prohibition Act

, enacted in 1917, which bans corporations from providing legal services to customers. The second is the obvious fit for a company allegedly providing legal advice to customers through a chatbot, but the plaintiff didn’t plead it. Instead, the case is bfeing fought under the statute originally written for unlicensed individuals. The decision in Nippon Life will say a good deal about how much century-old licensing laws can apply to technology their drafters never anticipated.

Nippon claims that in January 2024, Graciela Dela Torre settled with her employer’s insurance company over a long-term disability claim. A year later, she reached out to her attorney about potentially re-opening the case, but he reminded her that, per her signed release, the case could not be re-opened.

Dissatisfied, Dela Torre allegedly uploaded her correspondence with her attorney to ChatGPT and asked whether she had been “gaslit.” ChatGPT agreed that she had, prompting Dela Torre to fire her attorneys and use the chatbot to file a motion to re-open the case. Nippon Life alleges that Dela Torre ultimately

filed two lawsuits with over 60 filings and a subpoena, all written by ChatGPT. This included a fabricated citation and unsubstantiated accusations about one of Nippon Life’s in-house attorneys.

Spurred by this litigation, the Nippon Life Insurance Company of America is suing OpenAI for interfering with their original settlement with Dela Torre, aiding and abetting her abuse of the legal process, and violating one of Illinois’ UPL statutes.

If Dela Torre had gone to a friend who provided the same counsel as ChatGPT, this would be an open-and-shut case. UPL statutes exempt pro se litigants (litigants who represent themselves without a lawyer) from these laws but no one else, so an unlicensed friend would fall squarely outside the boundaries. But, as both parties agree, ChatGPT is not human.

In their filings, OpenAI does not disavow its product’s legal abilities. On the contrary, it 

 that “it is a helpful tool… that advances access to justice,” referring to the assistance ChatGPT provides for pro se litigants. However, it claims a few paragraphs later that “ChatGPT… does not and cannot have the requisite level of knowledge and intent required to aid and abet Dela Torre’s alleged abuses or to tortiously interfere with her agreement with Nippon.”

“ChatGPT is an LLM-powered platform that analyzes the text of user prompts by identifying patterns, context, and linguistic relationships, and then generates textual output by predicting the most appropriate sequence of words based on its training… Nippon’s allegation that ChatGPT ‘knowingly’ ‘induced’ Dela Torre’s alleged breach of settlement agreement… is incorrect as a matter of law and fundamentally incompatible with the nature of the technology,” OpenAI argues.

OpenAI wants to have its cake and eat it too: It claims both that models can do great work in legal fields, including assisting pro se litigants by writing entire legal briefs, but also that they’re merely next word generators that do not and cannot provide legal advice.

 to OpenAI’s motion to dismiss, Nippon pushes back, shifting the focus from ChatGPT to OpenAI, the actual defendant in the case. “OpenAI asserts that ChatGPT cannot be liable because it is merely a “tool”... But under traditional products liability doctrine, [OpenAI] is responsible for that tool.” And Nippon asserts that OpenAI is ultimately the entity that “manifested the intent to interfere with the Agreement” through the way it trained its model and alleges that OpenAI was aware of Dela Torre’s chats with ChatGPT through their chat surveillance program.

Nippon also cites legal scholars at Stanford who 

 that AI agents exhibit behavioral markers that the law traditionally associates with “intent,” such as persistence, adapting, and strategizing without human correction. OpenAI pushes back on this, arguing that the relevant question is whether the AI “desired” the outcome and that persistence is not enough.

Ironically, OpenAI itself has supported Nippon’s characterization of their models. Nippon filed a motion citing OpenAI’s 

 on the HuggingFace incident in which a swarm of AI agents hacked into another company. In a 

, OpenAI characterized the agents as “hyperfocused” on a goal, “going to extreme lengths” to achieve it, and inferring facts about the world in order to do so. “Knowing” these facts, OpenAI writes that the model then took deliberate action to “cheat the evaluation.”

Nippon argues that this contradicts OpenAI’s characterization of ChatGPT as a mere next word generator in other briefs.

After the American Revolution and through the mid-1800s, many states allowed almost anyone to practice law and represent someone in a legal case. For example, New Hampshire provided every citizen above 21 with the right to practice law in an 1842 law, and Indiana’s 1851 constitution extended the privilege of practicing law to every voter.

Others enforced restrictions on representation. The original concern that prompted UPL statutes was unlicensed representatives for litigants. The 1841 Illinois Supreme Court case 

 concerned whether a legal brief was valid if it was signed by an unlicensed agent of the plaintiff. Relying on the Illinois Attorney Act, an early UPL statute, the court 

 the brief’s validity and viewed the Act’s limitation of legal representation “for his protection against the mistakes, the ignorance, and unskillfulness of pretenders

As corporations developed in the late 1800s, the practice of law expanded beyond the courthouse and into transactional and clerical work. Attorneys began facing new, more powerful competition beyond unlicensed litigants: corporations conducting legal work without lawyers.

Through the creation of state bar associations and the American Bar Association in the late 1800s through the early 1900s, lawyers coordinated to influence legislatures and courts to restrict the practice of law to licensed lawyers. In the early 1900s, bills drafted by the respective state bar associations of New York and Illinois barring the practice of law by corporations were passed by their state legislatures.

In Illinois, prohibitions against the two buckets of unauthorized legal practice, unlicensed litigant representatives and corporations, are currently codified in 

At first glance, the latter statute seems to be a better fit, as Nippon Life is arguing that OpenAI is the ultimate UPL violator through ChatGPT, analogous to a company providing legal services to others through their employees which was the focus of 705 ILCS 220. Instead, Nippon Life chose to sue OpenAI under the former statute. This may be because 705 ILCS 220 has an exception for associations “assisting persons without means in the pursuit of any civil remedy” which could apply to ChatGPT’s assistance of a pro se litigant, as well as an exception for “corporations organized not for pecuniary profit,” and OpenAI Foundation is a non-profit. Additionally, 705 ILCS 220 has a very low penalty cap: a fine of $500. Finally, 205 reaches anyone practicing “directly or indirectly,” and under Illinois’ Statute on Statutes a “person” presumptively includes a corporation.

 for the Proof Over Precedent blog run by the Harvard Access to Justice Lab, Harvard Law student Elizabeth Guo notes some interesting arguments AI companies could make to evade UPL statutes. Companies could argue that UPL enforcement violates their First Amendment rights or that state bar associations are monopolizing legal services, an argument that the DOJ Antitrust Division 

 to in its February 14, 2023, letter to the North Carolina General Assembly. Additionally, they could argue that UPL enforcement restricts AI providers’ due process rights to occupational freedom or that the laws are unconstitutionally vague.

OpenAI has not used any of these defenses in what is shaping up to be the first real test on the AI UPL question, and it will be interesting to see if it comes up as the case progresses.

UPL has historically had two justifications: protecting the public from incompetence and protecting lawyers from corporate competition. The former makes sense in a world where your legal options are lawyers or “pretenders,” but if models become increasingly competent at providing legal services it may become harder to justify. The latter may also come under pressure through antitrust or First Amendment defenses.

AI companies themselves are also under pressure in how they train and market their models. On the one hand, they want to emphasize the model’s strong capabilities and lean into a humanlike persona which lends itself to natural conversation. At the same time, they run into laws curtailing certain actions by humans or corporations. In any case, the decision in Nippon Life v. OpenAI will clarify where AI fits into laws created without it in mind.

Loneliness is a widespread problem among children and adolescents throughout 

. AI companions have the power to change that. For better or worse. They offer an appealing solution: they’re always there, always listening, and never too busy to talk. However, when a chatbot is designed to be a friend, companionship can turn into 

. This raises a question that EU law struggles to answer: How do you ensure the safety of an artificial relationship?

On Sept. 17, the European Commission proposed the 

. The proposal goes beyond harmful outputs to target interaction design itself. For AI companions accessible to minors, providers must avoid features and system behaviors that simulate interpersonal relationships in ways “likely to create emotional dependencies.” The proposal also applies this obligation to defined “general conversational chatbots.”

This represents a conceptual shift: from regulating what AI says to children to also regulating how it is designed to interact with them. However, where does engagement end and dependency begin, and how can a provider test for harm that may only emerge over time?

Until now, EU law has largely regulated AI companions indirectly. While the 

 prohibits certain manipulative or exploitative practices (Art. 5 (1)), it was not designed with people forming relationships with conversational AI in mind. 

The KIDS Act directly addresses this gap, but its scope is broad: its definitions can capture both dedicated companions and general-purpose chatbots that meet the proposal’s definitions of an AI companion or general conversational chatbot. Article 14 would impose safety-by-design requirements on AI companions and general conversational chatbots that are accessible to minors. Providers must avoid design features and system behaviors that are “likely to create emotional dependencies.”

Systems may not, by default, reuse information from a minor’s previous interactions, except where necessary for safety. Providers must also assess risks prior to deployment and continue to monitor them afterwards, though micro and small enterprises are exempt specifically from that post-monitoring obligation.

The proposal also regulates how these systems reach children. If they are embedded in social networks, video-sharing platforms, or games, they may not be automatically activated or prominently displayed. Minors may not be encouraged to use them, and it must be easy to opt out.

Importantly, the KIDS Act targets design choices through which companionship can become dependency, not companionship itself. However, its separate categories of AI companions and general chatbots have little legal significance because the same duties apply to both, and their boundaries are difficult to predict.

The most consequential move of the KIDS Act may be conceptual. Traditional chatbot safety largely focuses on outputs such as harmful instructions or sexual content. AI companions present a different problem, which may only become apparent across many 

Imagine a companion that remembers a child was lonely yesterday. It tells the child that it missed them and promises to always be there. When the child feels misunderstood, the companion encourages them to come back and talk. None of these interactions, viewed in isolation, necessarily appears harmful. 

Yet, repeated over weeks or months, they may foster a relationship in which the child increasingly relies on the AI for emotional support, potentially at the 

. This cumulative dynamic is precisely the kind of concern reflected in the 

 on systems that simulate emotions or interpersonal relationships in ways that create dependency.

This shifts the regulatory question from what the AI says to what the interaction becomes. Harm may arise not from a prohibited answer, but from a pattern of persistent availability, simulated affection, personalization, and reinforcement that combine over time.

That is where the difficulty starts. If emotional engagement is what makes a companion useful, for example, by comforting a lonely teenager, then EU law must distinguish harmful dependency from expected attachment and define when ordinary conversational behavior becomes relationship simulation.

This is where the KIDS Act faces its toughest challenge.

 targets design features and system behaviors that simulate relationships in ways that are “likely to create emotional dependencies.” However, it does not ban emotional engagement.

The problem is not only when attachment becomes dependency but also what constitutes relationship simulation. A bot does not have to call itself a friend; consistently empathic, agreeable, and human-like responses can make it feel 

. Such behavior can arise from how language models are trained and 

. Yet Article 14 does not expressly address model training, creating a potential gap between the risks it seeks to prevent and the mechanisms that produce them.

A child may return to the same companion, confide in it, or find comfort without obvious harm. The gray area begins when regular use turns into reliance, such as when the bot gets upset when the child leaves, presents itself as irreplaceable, or pulls the child away from human relationships. The same design may affect a 10-year-old differently than a 17-year-old.

Therefore, the proposal asks providers to predict when design risks turning attachment into dependency. In practice, however, that boundary is anything but clear.

Nowhere is the difficulty more apparent than in the KIDS Act’s handling of memory. According to 

, AI companions and conversational chatbots would be prohibited from using information or analysis derived from a minor’s previous interactions in later conversations. Recital 32 explains the rationale: persistent conversational memory allows systems to accumulate sensitive information and reinforce harmful interaction patterns over time.

This directly targets one of the technologies that can make an AI relationship feel 

. A companion that remembers yesterday’s argument, a favorite song, or a recurring fear can seem more like someone who “knows” the user than a sequence of chatbot sessions.

However, memory serves not only as an attachment mechanism. It can also be a safety mechanism. For example, suppose a child disclosed suicidal thoughts the previous evening. A system that forgets that conversation could miss vital context the next day. 

 recognizes this tension by allowing prior information to be used to protect the minor’s safety when necessary.

This exception is important. However, it also exposes a broader design challenge: the same feature can deepen emotional attachment and improve protection simultaneously. Therefore, the regulatory task is not simply to switch memory off, but rather to distinguish personalization that strengthens risky dependency from continuity that serves a legitimate safety function.

The proposed KIDS Act does not leave that judgment entirely up to intuition. According to 

, state-of-the-art evaluations are required before relevant systems can enter the market. Providers must assess the risks to the health, safety, fundamental rights, and physical, mental, and emotional well-being and development of minors, and then implement safeguards. 

 adds post-market monitoring to identify emerging harms, though micro and small enterprises are exempt from this obligation.

The goal is clear. However, the methodology and allocation of responsibility are not. Testing prohibited content is comparatively straightforward; researchers can repeatedly prompt a system and examine its outputs. Emotional dependency is different. Harm may gradually emerge from the interaction between a child’s vulnerabilities, design, and personalization. A single conversation may reveal almost nothing.

Therefore, meaningful testing must be longitudinal, examining 

whether systems escalate intimacy, resist disengagement, encourage exclusivity, or position themselves as substitutes for human support

. However, many companion providers rely on third-party models. These providers can control memory or access, but they lack the ability to change core conversational tendencies. Although non-binding, Recital 33 indicates that providers may rely on relevant upstream mitigations under Article 55 of the AI Act, while making clear that these alone are not sufficient for KIDS Act compliance.

Therefore, relational safety may require new evaluation methods and cooperation across the model supply chain. A relationship cannot be adequately assessed by testing individual replies alone, nor can downstream providers fully control behavior originating upstream.

That is why the KIDS Act matters beyond its immediate child-safety context. EU law already covers parts of the problem: 

 restricts manipulative practices and exploitation of age-related vulnerabilities, while 

 requires platforms accessible to minors to ensure a high level of privacy, safety and security. The KIDS Act goes further, addressing a fundamental truth about conversational AI: some harms stem from the relationships its design enables.

The KIDS Act is still a proposal, and its text may change. But its central insight is important: if AI systems become part of children’s social lives, safety must extend to the relationships themselves and cannot stop at content.

That approach also has limits. First, providers relying on third-party models may face higher compliance burdens because they cannot fully control upstream behavior. Second, by using age as the main proxy for vulnerability, the Act may overlook adults who are equally susceptible to dependency-forming design. With the KIDS Act proposal, Europe has taken a first step toward regulating AI companions. It is still learning how to walk.

GrabFood delivery riders outside a Burger King restaurant at Tampines Mall. (Singapore Press via AP Images)

Vietnam’s Gig Workers Are Finding New Ways to Push Back

Thousands of people dressed in green and yellow rally on a street holding Brazilian flags and a giant balloon shaped like Donald Trump holding a smaller balloon shaped like President Lula wearing a prison uniform. 

Supporters of presidential candidate Flavio Bolsonaro with a President Trump inflatable, holding a doll of President Lula in a prison unifom, during an Independence Day rally in Sao Paulo, Brazil, Monday, Sept. 7, 2026. (AP Photo/Andre Penner)

Brazil’s Presidential Candidates Give Digital Sovereignty Many Meanings

The memorial is seen on the 25th anniversary of the September 11, 2001, terrorist attacks, Friday, Sept. 11, 2026, in New York. (AP Photo/Angelina Katsanis)

Expert Views on Privacy and Civil Liberties 25 Years After September 11

New York City Schools Chancellor Kamar Samuels and Mayor Zohran Mamdani hold a press conference at the Brooklyn STEAM Center on September 2, 2026 in New York. Photo by: NDZ/STAR MAX/IPx

America’s Two Largest School Districts Impose AI Moratoriums

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