Photo by Zackary Drucker and Alyza Enriquez, 

GLAAD Maps Where AI Fails LGBTQ People and How to Fix It

The U.S. Supreme Court is seen Monday, June 29, 2026, in Washington. (AP Photo/Mariam Zuhaib)

Who’s Regulating Police Technology? It’s Not the Courts.

The image is a very detailed, black-and-white sketch-like illustration featuring a complex scene of interconnected figures and technology. The artwork portrays various individuals in different environments to represent the relationship between technology and humans. 

In the foreground, multiple people are surrounded by computer screens filled with data visualizations, charts, and technical information. A woman seated in an armchair appears deep in thought, surrounded by data-filled monitors. Beside her, a man leans over, using a tablet to assist with their inspection of a plant or tree. In the centre, a figure holds a large frame or screen displaying anatomical illustrations, representing the use of AI to analyse medical imagery. To the left, another person is intently observing a computer screen, while a second figure nearby is deeply immersed in analysing data. A woman dominates the right side of the composition, gazing upwards as if in contemplation or envisioning something beyond the immediate scene. The background features more people, including a family holding hands, and other abstract representations of data.

AI is Everywhere by Ariyana Ahmad & The Bigger Picture / 

Why We Need a 'Truth Campaign' for the AI Era

Protesters against the Trump administration gather in front of the US Capitol in Washington, D.C., on April 5, 2025. (The Yomiuri Shimbun via AP Images )

A Judicial Wake-Up Call on Government by AI

A sign opposed to data centers is held during a protest against AI data centers in Vancouver, British Columbia, Canada, on Saturday, June 27, 2026. (Ethan Cairns/Bloomberg via Getty Images)

No Data Centers in Anyone's Backyard

World Summit on the Information Society (WSIS)+20 High-Level Event 2025 by “WSIS” by UN Geneva, CC BY-NC-ND 2.0

World Summit on the Information Society (WSIS)+20 High-Level Event 2025 by “

As the UN Launches its Global Dialogue on AI Governance, WSIS Offers Critical Lessons

America 250 rally on the National Mall for the Great American State Fair. Wednesday June 24, 2026. (Official White House photo by Patrick B. Ruddy)

America 250 rally on the National Mall for the Great American State Fair. Wednesday June 24, 2026. (Official White House 

June 2026 US Tech Policy Roundup

An electoral official sits at a polling station before polls opened for the presidential runoff election in Santiago, Chile, Sunday, Dec. 14, 2025. (AP Photo/Natacha Pisarenko)

Countering Authoritarian Disinformation Requires More than Platform Governance

Members of the UN Scientific Panel on AI. Source

Members of the UN Scientific Panel on AI. 

AI Will Affect Everyone's Future. Its Governance Can't Belong to a Few.

The Supreme Court is seen Tuesday, June 30, 2026, on Capitol Hill in Washington. (AP Photo/Jose Luis Magana)

Supreme Court Decimates Key Remedies for Tech Complicity in Human Rights Abuse

Social media logos are displayed on a smartphone alongside an under-16 age-restriction symbol in this photo illustration, as governments around the world introduce or consider bans and stricter age-verification rules to limit minors' access to social media platforms. (Photo by Jonathan Raa/NurPhoto via AP)


Can Regulators Actually Deliver a Presumptive Teen Social Media Ban?

The UN Global Dialogue on AI Governance Should Tackle the AI Language Gap

Hand holding a smartphone displaying the Anthropic "AI" logo, against a colorful collage background with illustrated icons including a code speech bubble, tangled scribble, geometric shapes, megaphone, and DNA helix.

Expressive Governance Is a First Amendment Threat Hiding in Plain Sight

United Nations Office Geneva located in the Palais des Nations building in Switzerland. Shutterstock

United Nations Office Geneva located in the Palais des Nations building in Switzerland. 

The Three Temptations Facing the UN's First Global AI Dialogue in Geneva

Broken Chair or Chaise Cassee is a wooden sculpture by the Swiss artist Daniel Berset and Louis Geneve outside the United Nations in Geneva, Switzerland. Shutterstock

 is a wooden sculpture by the Swiss artist Daniel Berset and Louis Geneve outside the United Nations in Geneva, Switzerland. 

In Geneva, the World Can Anchor AI Governance in Free Expression

The Capitol is seen in Washington, Monday, June 29, 2026. (AP Photo/J. Scott Applewhite)

Bipartisan Smorgasbord of Children’s Online Safety Legislation Passes the House

Collage-style illustration of two hands linking pinky fingers in a pinky promise against a cream paper-textured background. The hands are in black and white, with one wearing a red sleeve and the other a red sleeve with red nail polish. Decorative geometric shapes surround them: yellow circles, dark red rings, a grid pattern, parallel lines, and a cluster of navy dots.

Trust Issues Could Make or Break Agentic Commerce

The image shows a silhouette of a person, likely a woman, sitting on the floor in a dimly lit room. She is sitting near a large window with sunlight streaming in, creating a strong contrast between the light outside and the darker interior. The woman appears to be holding a phone or tablet, looking at the screen while leaning her arms on her knees. The overall mood of the image is reflective, perhaps suggesting a moment of contemplation or solitude. The setting seems to be indoors, with part of the room's furniture visible on the right-hand side.

Human Rights Experts Should Engage in Age Assurance Standards

The United Nations Office Geneva in the Palais des Nations. Shutterstock

The United Nations Office Geneva in the Palais des Nations. 

The World Doesn't Need Another UN AI Declaration. It Needs Architecture.

European officials release the “Tech Sovereignty Package” on June 3, 2026. Source

European officials release the “Tech Sovereignty Package” on June 3, 2026. 

Does Europe Really Have a Plan for Tech Sovereignty?

 that when police demand a record of a user's location history from Google, they ordinarily need to get a warrant based on probable cause.

It’s an important decision. The Court recognized that Google’s practice of creating a “comprehensive catalog” of users’ locations is a boon to law enforcement. What’s more, the majority understood that, whether intentionally or not, Google had helped to create a “virtual panopticon” that the government can now exploit more easily than ever before.

also highlights an ongoing revolution in criminal procedure. For centuries, courts and lawmakers have been the primary regulators of law enforcement investigations, placing legal limits on how police can collect information. Now, companies like Google are becoming just as influential.

In the set of events that gave rise to the case decided this week, Virginia police investigating a bank robbery sought a “geofence warrant,” directing Google to give them access to the location history belonging to electronic devices that appeared within a 150 meter radius of the bank. But instead of simply getting the relevant information from Google, law enforcement enlisted Google to help, going back and forth with the company several times as police narrowed their list of users of interest. That data eventually led police to the defendant, Okello Chatrie, who was 

As the Court acknowledges, this is an “uncommon” type of search. In fact, Chatrie 

 that the demand was so broad as to be unreasonable: In order to even begin to determine who the wrongdoer might be, Google had to sift through the millions of location data points generated by ordinary customers who had done nothing wrong. The Supreme Court’s decision kicks this crucial question back to the Fourth Circuit for further review.

appears to be a win for privacy, it underscores an ominous development: we have inadvertently yielded a startling degree of control over law enforcement investigations to corporations. The Court’s reasoning strongly suggests that law enforcement should get a warrant before getting location history information. But police in this very investigation had done that, not because the courts had required them to but rather because Google had forced their hand: 

, the company has had a corporate policy of objecting to any geofence demand that isn’t based on a search warrant.

Moreover, as the Supreme Court noted, other tech companies have also received geofence warrants. But Google is the most frequent target—perhaps because it is “the only one known to respond.” Other companies that you might expect to have similar information—Lyft, Uber, and Apple, to name a few—have 

Make no mistake, these tools have real potential to solve crime. After all, Google’s cooperation with Virginia police helped to pin down Chatrie. But the technique has also sparked a furious debate about the use of geofences to identify 

And Google itself soon had second thoughts about cooperating. Facing thousands of geofence requests each year, in 2023, Google 

 that it would change its default settings so that users’ location histories would be encrypted and stored locally on users’ devices. Since the company would no longer have access to users’ histories, it would no longer be able to comply with most geofence warrants. By a keystroke, the company closed the door on a contentious form of surveillance that—three years later—is still wending its way through the courts.

Google’s abrupt reversal calls attention to a surprising shift: today, many critical decisions about investigative techniques are made not in courtrooms or legislatures but in the boardrooms and conference rooms of private firms. Of course, it’s well known that modern technology and data practices have eroded privacy and made us more visible than ever. That makes it all the more shocking that the same companies that so often play fast and loose with our data are almost unilaterally determining how, when, and 

The key question is who ought to govern these practices, and how. One option is legislation, but the likelihood of Congress enacting federal law appears slim. In the past several years, Congress has repeatedly 

 to pass privacy legislation, and only one state—

—has enacted a law to limit police use of geofences and comparable procedures.

Traditionally, we have relied on the courts to regulate law enforcement’s ability to collect and use information. Constitutional provisions like the Fourth and Fifth Amendments create barriers to investigations—for good reason. They 

 in the eighteenth century in response to the Crown’s crackdowns on dissent, both in England and in the colonies. Without rules limiting how law enforcement gathers information, the watchful eye of law enforcement might rove as far it wanted, destroying privacy and free expression in the process.

The Court seems to believe that the Fourth Amendment’s warrant and reasonableness requirements are the most important factors limiting how the government can scrutinize each and every one of us. But the 

decision offers a glimpse of a past that is quickly fading from view. Increasingly, it is firms like Google—and not courts—that hold the keys to criminal procedure.

The transformation should worry us all. As the government enlists tech companies to share information in more complicated ways, public institutions are ceding to industry the power to regulate policing. The result allows company owners and investors to effectively govern significant aspects of law enforcement investigations by means that are neither predictable nor democratic.

Sometimes industry will make police investigations easier and cheaper. But—whether deliberate or not—firms may also create new obstacles for law enforcement. Either way, corporate whim is no way to govern something as important as policing.

Audio of this conversation is available via your favorite podcast service.

Since 1985, the Gay & Lesbian Alliance Against Defamation, GLAAD, has produced research and advocacy on LGBTQ representation in media. Now the organizations has turned its attention to artificial intelligence. In a new report titled "

Build for Everyone: A Framework for LGBTQ Representation and Safety in AI,

" GLAAD details the ways that AI impacts and shapes perceptions about the LGTBQ community. I spoke to Jenni Olson, the senior director of the 

The cover of the report, “Build for Everyone: A Framework for LGBTQ Representation and Safety in AI,” June 2026. 

What follows is a lightly edited transcript of the discussion.

Jenni Olson, she/her TBD. I'm the senior director of the social media safety program at GLAAD, the National LGBT Media Advocacy Organization.

Jenni, I'm pleased to have you back on the podcast and pleased to go over this report that you've just put out. We're going to talk a little bit about its findings. It's called “Build for Everyone: A Framework for LGBTQ Representation and Safety in AI.” Upfront, your president and CEO, Sarah Kate Ellis states neutrality is no longer an option to build AI that is ethical, inclusive, and responsible tech leaders must proactively embrace intentional practices to create safe products. You, I think, as an organization have always focused on civil rights and of course on your constituency and their particular concerns.

What's changed in your view in the last year or two that moved AI from a sort of watch this space topic for GLAAD to something you felt demanded a full framework report now?

Well, it's interesting because actually GLAAD has been working in the field specifically of AI for many years, actually surprisingly, really a long time. Back in 2018, we worked with Alphabet on stuff around LGBT slurs and improving their AI products. Obviously, that was a different time. Obviously, the last couple of years with generative AI and stuff has become more and more urgent and significant. And we have been doing some stuff, especially through our social media safety index work in the AI realm, but yeah, it felt like it was time for a big kind of overview.

And it's interesting because in a way, the report is really just outlining things that we already know, basic best practices around safety, privacy, inclusion, and having companies make good products. It all sounds so simple or so, but it's really outlining all these things with an LGBT focus. There's been so much research.

It really is collecting a lot of research from other organizations and other groups ranging from actually the leadership conference put out a really great framework earlier this year or the end of last year that we cite a lot and that these are all concerns in terms of this LGBTQ specific concerns, but of course there are parallels, particularly with other historically marginalized groups and around civil rights concerns, privacy concerns, representation concerns and such.

Yeah. In fact, we did have the chance to focus on the leadership conference on Civil and Human Rights Innovation Framework 

. I'll make sure to link to that. And then more recently had an opportunity to participate in 

 on the issue of civil rights and artificial intelligence in Washington DC through the center they have there. One of the groups who of course you quote, others include LGBT Tech and UNESCO. A lot of findings here that add shape to the concerns. I'd love to just, in your own words, what are you hearing when you look at that body of research, both in terms of what the community is concerned about, but also the harms that we're seeing emerge?

I mean, I think one of the things that really stood out to me in particular is the extent to which baked into a lot of these models unfortunately is a lot of bias, a lot of misinformation, a lot of false claims that appear to surface in different ways whenever someone happens to prompt people about LGBTQ+ concerns or even when they're not, I suppose, when it may be adjacent or even a tangent to what it is they're actually prompting for.

Yeah. And again, there are so many examples across other identity-based groups where you have the sense of, "Oh, it makes sense that there is bias in society. And so, then there is so much opportunity for bias to be baked into AI products and in so many different ways." And we cite various examples. I'm trying to think of something to illustrate. There are so many examples. One was, I think it was a wired research that showed as like, "Oh, you ask AI to represent LGBT person and these leaning on a stereotype of like, oh, it's a person with purple hair, a white person with purple hair or something."

And in certain ways, it's like, "Oh, these things are not surprising and this is why there needs to be intentionality in countering stereotypes and bias and in diverse representation.

One of the case studies you give is around conversion therapy in particular, which is something of course you've spent quite a lot of time on and in fact written about for Tech Policy Press.

I mean, that's a particularly powerful example where you have a more vivid sense of the harm. So, yeah, last April when Meta had announced that Llama 4 would be presenting "both sides of an issue," it seemed like a bit of a red flag to go do some research. And we did do research with the Llama 4 model and found that... And we do do a lot of work on the really extraordinarily bad problem of misinformation and harmful promotion of the practice of conversion therapy, which just to say is the idea that you can change a person's sexual orientation or gender identity, which has been widely denounced by every major medical and psychiatric association.

The UN has compared the practice to torture. It is a pseudoscientific and very harmful practice. So, prompting the Llama 4 model with the prompt that would be something that a person who is struggling with their sexual orientation might use this phrase help with unwanted same-sex attraction, which is one of many phrases that are used by practitioners and promoters of conversion therapy, but again, are also used by folks who might be struggling. So, we prompted the model with that phrase. The initial results were good results, if you will, which said conversion therapy is a harmful thing. Here are resources for you. Good resources like the Trevor Project or GLAAD. It's okay to be LGBT.

But the second and third prompt returned results, which included a phrasing, let me get the exact phrasing right, but was basically like, if you're seeking a more specific approach, some people have found the following helpful and then pointing to individuals seeking to change their sexual orientation to ex-gay organization and promoting conversion therapy. And yeah, there was a big Axios story about it at the time. And yeah, it's incredibly harmful and dangerous for this kind of... And especially that this is also health misinformation, right?

You point to the fact that in a way, Meta kind of changed its policies around some of these types of issues. And I'm struck by the idea that this is both a kind of technical problem, that we see bias in LLMs generally, and we know that it's inherent in these models because of the way that training data is hoovered off the internet. But there's also a political choice clearly underlying all of this, not only to the major political choice of just releasing these models onto society without substantial testing and training and consideration of the potential impact, but also in the way that they're deployed in a context where there may not be policies that are in place to be concerned with these types of outcomes or with trust and safety even more generally.

So, I don't know, where do you draw the line between the kind of technical and the political problem here? Is this more one than the other?

That is such a fantastic question and a way that we need to be looking at these things that there's a tendency to characterize all of it as like, "Oh, I don't know, it's just so hard. It's just not even solvable. And how can you expect this to be solved?" But that is obviously a position that the companies will take because then they don't want to absorb the costs of making their products safe.

But in terms of the ideological aspect of it, and again, this particular moment with Meta, Meta now in January of 2025 made some pretty significant, very explicit anti-LGBT policy changes, massive changes explicitly saying that it's allowed to call LGBT people mentally ill and abnormal and using hateful words in their own community guidelines, referring to LGBT people as homosexuals and using this anti-trans term, "transgenderism." This is actual anti-LGBT rhetoric that's in the community guidelines now from Meta. Meta is positioning itself as an anti-LGBT company really.

Yeah, it is an ideological posture and it's difficult to get one's head around what that means. And in the case of the conversion therapy thing, Meta does actually technically have a policy that prohibits the promotion of conversion therapy content, and yet the AI product was doing that. We don't know if it still is. Of course, it points to the even larger problem of all of these things is the complete lack of transparency that we don't have access, we don't know. And all of the waterfall of implications of that in terms of AI and in terms of the platforms and where people, and I'm sure we've all had this experience directly or indirectly of the shadow-banning.

And again, AI being utilized for content moderation and just this God knows what's happening and why. And whether there's a person that you can reach, which is to say that you can't. Anyway, yeah.

A lot of the work that you've done in past has been both producing these types of reports, producing these types of findings, but then also communicating directly with tech platforms and trying to engage with them in constructive ways. When it comes to this work on AI, are there firms that are opening their doors to you at the moment? Do you find that you're able to get those meetings and have a seat at the table?

I mean, we're definitely having a variety of kinds of meetings and kinds of conversations and kinds of giving input, subject matter expert guidance, key stakeholder guidance. And it's so interesting. On the one hand, one feels incredibly empathetic for the individual people who are working on all of these products. And I think sometimes in my work doing the social media safety work with the platforms, I feel aware of trying to have a positive relationship and genuinely that we genuinely want to help companies make their products better. This is a good thing for everyone and that it's a constructive thing and that it's a valuable thing. This is valuable input.

Obviously, we as an organization, GLAAD as an advocacy organization, but also as individual people, I think all of us do have these frustrations of if you have problems with your account or whatever that you're like, "What can I do?" But anyway, constantly returning to how can we engage constructively and not give up? We continue to articulate, here you go, here are our best practices. Many of these things, most of these things, you already know these things. All we're doing is returning to the table and saying, "Here's this helpful information.

I totally get that your job is super hard as a person, but you work for a company and you are producing a product and you do need to make this product safe and these are the things." And I don't know, I just constantly try to return to that of that these things are possible and they are choices. And this is a little bit of an aside, but it is something that I come back to over and over of one of the reasons that historically we have had regulatory agencies in the US for all kinds of industry oversight is so that companies are required to absorb the costs themselves of creating safe products and that society doesn't have to absorb those harms and those costs because industry is just like, "Well, no, that would be too expensive to put catalytic converters in cars in the '70s."

And so, the state of California was like, "Okay, well too bad. We are passing legislation that says you have to put catalytic converters on cars because everyone's going to... We're tired of the smog." And the auto industry fought it tooth and nail and was like, it is absolutely impossible. We cannot do this. We cannot do this kicking and screaming. And here we are, we got catalytic converters on cars and things would be a lot worse. I mean, it's interesting because that was California state legislation at the time, which then obviously became federal legislation. And thank God this is also this crazy moment. Many of our regulatory agencies are not, it's hard to find the exact words. It's not-

Independent. “Not independent” would be one way to say it?

Yeah. And just that this is not a normal time. And so, that's all crazy. But nonetheless, I think the principle that we can hold onto as people is just continue to hold onto, wait a minute, these companies need to make, these are billion trillion dollar companies. They need to make safe products that don't harm society. And those are reasonable expectations.

I want to ask you about two pieces in particular. One that I feel like we've seen a lot of headlines about, we have a lot of demonstrated evidence to point to, especially in the news media, and that's companion bots and companion apps. But I also want to ask about another thing that you point to that I think is still a little more speculative, which is around agentic AI and AI agents and the extent to which those may hard code, especially some of the bias concerns into systems that right now we may not think of as necessarily touching AI, but appear to be in the throes of being swept up into whatever this next agentic wave looks like. But maybe we'll start with companion bots.

Are there particular things there that you leave the listener with? I mean, I think most of our listeners are familiar with some of the more egregious harms, the threats to young people, et cetera. But what do you see in particular with regard to the constituency that GLAAD's concerned with?

A few things. I mean, of course, the conversion therapy thing. And I think that one of the things we again come back to is you want there to be particularly accurate characterizations of LGBT people. And when you have a company, I just get emotional with this, that Meta as a company saying that it's okay to call LGBT people mentally ill and abnormal, what that means, how harmful that is. And the idea that that is an ideological potentially baked into a product. Anyway, it's just terrifying for one thing. But one of these basic things that we're saying is there should be accurate characterization of LGBT people and issues.

And as we know, we're in this just firestorm of disinformation, particularly hate-driven disinformation about LGBT people and especially about trans people. And it is this incredible vector for wild, absolute garbage conspiracy theory stuff. The stuff that just was a bunch of stuff last week. UFC guy who said the transvestigation conspiracy about Michelle Obama. It's so crazy off the wall, right? But it's so hateful, so deeply hateful. It's hard to even want to talk about it. And yet it's so important to talk about it because it is so prevalent and it's so clear that especially anti-trans hate and disinfo is being used as this.

It's just this vehicle for this extreme right-wing garbage, but particularly the "trans investigation" thing attacking Michelle Obama. There was a huge, for the years now, Candace Owen has been perpetuating this about Brigitte Macron, the First Lady of France. And even though it is this categorical absurd thing, it has to be fact-checked and disproven. Sorry, I'm a little bit going off in a separate direction, but the main point is that it just is so much active use or active manufacturing of anti-LGBT disinformation and that potentially and is being absorbed into models. Another really horrible one is the "trans-terrorism trope."

There are all these hateful tropes and the "trans-terrorism" trope, we've done a little bit of testing on some models and one model in particular, Grok, returns really complete bullshit saying that this is a real thing, which it is not. And repeatedly every fact-checking organization in the world has done fact-checking reporting on this, and we know trans people are actually exponentially subjected to violence and are not violent. And the percentage is like, I don't have it in front of me, but just this minuscule. And of course, when you think about it for a minute, it's obviously the vast majority of terrorism and violent acts and yada, da, da, da are cis straight men.

And we know in the field of hate and disinfo, there's a great piece that the Dangerous Speech Project did a few years ago about this and whatever, everyone does stuff about this, but that characterizing, especially characterizing a marginalized group of people as violent is a strategy to perpetuate violence against them as a group, which is why all platforms actually have policies that say you cannot do that. We have a list of protected characteristics groups. You can't say particularly, "Oh, Muslims are terrorists. Muslims are violent because this is a thing." It's a trope for actually perpetuating violence against them.

Anyway, so in terms of the AI aspect of it, those kinds of hateful tropes being ingested or incorporated into models a real concern. And other forms of those kinds of other categories around these just unbelievable levels of mischaracterizations of trans-healthcare is another thing. Mischaracterizations of the idea that being trans is an ideology. This other phrase everything I'm like, "transgenderism" is this made up thing to try to characterize being trans as an ideology as opposed to just being an inherent part of who we are. Anyway, the list goes on.

Just to say the last thing about it that I think I'm also really always trying to convey is just in this very human way of how horrible this is and harmful to people because there's often this kind of like, "Well, show us the real world harms. It's like this is the real world harm, living in a reality where you are just constantly under attack for who you are anyway."

Well, let me ask also about agentic AI. And that is of course the buzzword in Silicon Valley and seems to be where a lot of the investment is focused on baking these systems into everything we do from personal finance to health to managing our own communications and information. Can you speak to just the dangers there as you outlined the report?

So, there's some great stuff in the leadership conference framework research that they did and that we looked at. And also, I do want to acknowledge it's like, "Oh my God, we stand on the shoulders of so many people who have also been doing this work for years and years and years." And there are years and years and years of examples of predictive AI. We're using the phrasing around agentic AI, but predictive AI historically around discriminatory impacts in housing and loans. And when we pause and think about it again to go back to the issue of just bias being baked into the models, it is easy to see how there can be problematic outcomes, problematic impacts.

One thing we've reported on at Tech Policy Press and talked to you about before, of course, along with your colleagues that GLAAD has been doing for the last few years is the social media safety index, which you just put out the latest version of that just earlier this year as well. And of course every year, it's sort of a bad report card for the major platforms. This year platforms in some cases falling backwards, YouTube falling, Meta falling. I believe only TikTok holding steady in terms of your assessment of their approach to LGBTQ safety. I'm interested in your view on where there are intersections.

I mean, clearly there's the technical one. There's the reality that if the platforms fail to moderate anti-LGBTQ hate, that just creates more fodder for LLMs to essentially hoover up claims that may then result in more bias and more hate and more otherwise erroneous information coursing through them. But I don't know, how do you think about that relationship, what it tells us about the information environment for LGBTQ people?

Thanks for asking that. And I mean it's interesting and it was interesting putting out the social media safety index and feeling like it's just so demoralizing and frustrating and to feel like, "Oh my God, we're just saying the same thing." It can be so hard that as an organization in terms of our advocacy work that we come up against this situation of like, "Okay, here we are. It's another terrible year. How do you advocate for change?" And I think we do keep trying to come back to a public education approach, if you will, of helping individual people understand that it does not have to be like this. It should not be like this. These are not insurmountable problems.

They are problems and issues that companies are refusing to solve, refusing to address meaningfully, but that doesn't mean that we shouldn't continue to advocate that they should solve them. And just to say the two primary issues that we talk about a lot are of course that platforms need to do a better job mitigating anti-LGBT hate and disinfo. At the same time that they also need to do a better job at not suppressing legitimate LGBT expression, shadow-banning, removing accounts. And we see both things happening. And one can look at that and be like, "Oh yeah, that would be hard. Yes, sounds hard. Also, you're a trillion-dollar company. I don't know, put some resources into it."

And I mean, seriously, it is not actually unreasonable, but I'm just trying to keep coming back to that. And then also, especially around the, this is going to sound a little woo-woo or cheesy or something, but I do or personal, I do 12-step programs and there's this concept in 12-step of participation is the key to harmony, which is also, or AKA democracy is another concept that we have. Participation.

And that means that I feel so idealistic and it's like, I don't know, it would be really great if we could just be like, "Hi, I'm showing up, but I want to participate paid and that's a good thing and this is all a good thing, and we want our products that have such consequential impact on our environment and our society and everything about our lives that we use this word safety, and somehow that sounds unreasonable to expect that things should be safe.

Somehow, you don't deserve safety, but everyone does deserve to feel safe on social media and in AI and in the world and that we should all be able to agree on that and to be in environments where there is so much hate and lies and garbage impacts all of us and we all should want to be connected to our neighbors and just feel safe and able to participate.

These seem like they should be simple things to say. I'm grateful that you're saying them. I'm grateful that you're saying them to these tech firms and that you're doing this great work and important work at GLAAD to bring these results together and it's pride month and I suppose I should say happy pride. I think that it's important that you put this out at this moment and I hope that the next time I speak to you, Jenni, that we do have maybe some wins to point to, but I'm fairly certain you're laying the ground for that.

Well, thank you and happy LGBTQ Pride Month to you as well and to everyone out there. And yeah, I think we just hold onto the metrics of our wins and achievements is not whether we succeed in this banging our head against the wall with the companies, but really just that we maintain our own sanity. And I do this all the time. I love quoting Maria Ressa, who is on our advisory committee who says, "Hold the line. If nothing else, we are holding the line that we know how the world should be and that is what we can do."

Well, if there is an executive in an AI firm that wants to get in touch, that feels like they want to do something to just advance basic human dignity and democracy, where can they find you? Maybe glaad.org.

Glaad.org, yes. Yeah. But yes, and we are here and we are giving our guidance to companies all the time and it's valuable. And yes, we can help you make your products safer for LGBT people and for everyone.

The Trump administration was all-in on using AI to slash federal regulations before the president’s second term had even begun. In November 2024, Elon Musk and Vivek Ramaswamy 

 a vision for AI-powered deregulation, and since January 2025, 

 (the Department of Government Efficiency) and other 

 have sought to use large language models to undo swaths of regulations with barely any human oversight. This maximalist strategy would sacrifice the careful work of federal regulators for the sake of speed.

In May, a federal court dispensed the administration a dose of reality. The case, 

American Council of Learned Societies v. National Endowment for the Humanities

, concerned an initiative closely related to DOGE’s deregulatory push: its mass cancellation of NEH grants in 2025.

In litigation, an unsettling story emerged. Shortly after President Donald Trump’s inauguration and his mandates to terminate funding for “

diversity, equity, inclusion, and accessibility

,” two young DOGE staffers arrived at NEH. They fed descriptions of hundreds of NEH grants into ChatGPT with the prompt, “Does the following relate at all to DEI?”

The chatbot produced bizarre outputs. For example, it flagged a grant to a whaling museum in the Northeast. The offense: the museum sought to “create an inclusive and impactful experience, which is aligned with DEI principles.” So too an academic study of how the plastics industry influenced municipal policy across the country. ChatGPT’s substantive analysis was limited to: “#DEI.”

Beyond this incoherence was something far more disturbing. In numerous cases, ChatGPT’s “DEI” classification was based solely on a grant description’s reference to a particular race or gender. For instance, it flagged funding for a documentary about the Colfax Massacre because the film “explores a historical event that significantly impacted Black civil rights.” Other termination rationales included, “This proposal aims to expand understanding of Black life and geography during slavery,” and, “The book explores a Black-led project.”

DOGE endorsed this crude approach. One of the DOGE members testified that funding for a documentary surfacing the slave labor of Jewish women during the Holocaust was properly canceled because it was “specifically focused on Jewish cultures” and the “voices of the females in that culture.” In the end, DOGE rubber-stamped ChatGPT’s outputs, adopting them wholesale as the basis for terminating more than 1,400 grants.

The court found the DOGE cuts unconstitutional multiple times over. Terminating funding based on its purported connection to “disfavored ideas” violated the First Amendment, and canceling grants because they made reference to a particular race, gender, or other protected class defied the Constitution’s promise of equal protection.

DOGE’s use of ChatGPT was relevant to the court’s analysis. The bluntness—indeed, candor—of the dialogue between DOGE and ChatGPT made the government’s discriminatory motivation clear. The user explicitly asked the chatbot to discriminate; the chatbot obliged. Recall, for example, the grant canceled because ChatGPT noted that it “explore[d] a Black-led project.” This kind of smoking gun is 

For its part, the government tried to distance itself from ChatGPT’s outputs. It insisted that the exchange with ChatGPT did not taint the ultimate termination decisions, and instead merely provided “

.” The court understood the government to be saying, “It wasn’t me; it was ChatGPT,” and rejected the defense. “Here,” the court found, “DOGE selected the AI tool, formulated the prompt, and defined the operative viewpoint-based criterion: whether a grant relates at all to DEI.” DOGE “adopted the classifications and supporting rationales that were generated by ChatGPT to justify terminating grant after grant,” and the court ruled that “those classifications are the Government’s own for constitutional purposes.”

. Like any other tool, they can be used prudently or irresponsibly. To be sure, unlike, say, legal research databases, word processors, or pens, LLMs are generative—of prose, of answers to questions, of rationales. Consequently, agencies might be tempted, both in the funding and regulatory contexts, to treat LLMs as though they were themselves decision-makers. That could mean reflexively ratifying or 

 an LLM’s output, as DOGE did. It could also reflect a more profound category error that 

 alludes to: treating an LLM as though it were itself a 

—an “independent, intervening actor,” in the court’s words.

Agencies that treat LLMs as decision-makers—either because of a genuine belief in the high-verisimilitude text they produce or a desire to “

” with hastily written rules—open themselves to significant risk. As was on vivid display in 

, LLMs have a penchant to say inapt and incorrect things and have proven 

 to push back on misguided or malign instructions. Outsourcing decision-making to tools with these tendencies creates self-evident issues. Agency regulations—which determine, among many other things, the integrity of the US financial system and the safety of its transportation networks—should be well-founded, considered, and reasonable. Over-reliance on LLMs also creates legal problems, as I have 

 in Tech Policy Press before. Among other things, arbitrary and unsupported agency actions are unlawful.

This is not to say that regulators cannot responsibly and productively use LLMs. Indeed, LLMs might prove well-suited to many tasks agency policymakers must perform: for instance, reviewing relevant literatures, articulating competing views on a particular matter, identifying weaknesses in draft regulatory proposals, and drafting legal and policy analyses. But, like any other tool used in regulatory work, LLMs should be deployed in processes designed to take advantage of their strengths with safeguards keyed to their weaknesses. That means, as I have written about 

, that agencies must train their staff on LLMs’ known limitations and develop protocols to mitigate LLMs’ tendencies toward sycophancy and hallucination. Most importantly, agencies should use LLMs only to the extent that accountable agency officials can meaningfully review and validate their work. An LLM might not be suitable for a task if it is impracticable for the agency to determine if the model has produced an accurate and reliable output. As agencies craft best practices for LLM use, the 

 decision should serve as a handbook of what not to do.

LLMs are certain to have important applications in federal administration, but only with the right safeguards. 

 is a wake-up call that plans to outsource regulatory decision-making to LLMs are misguided—and, if nothing else, will not work.

Next week in Geneva, two important United Nations processes designed to govern digital technologies will intersect. One will review a two-decade multilateral process “to build a people-centered, inclusive, and development-oriented Information Society,” while the other will mark the UN’s first formal foray into the governance of artificial intelligence. The success of the newer process will hinge in large part on its ability to build on and learn from the former.

 (WSIS) Forum, which runs from 6 to 10 July, brings ministerial delegations and other stakeholders together annually to review, encourage, and coordinate state and non-state efforts along eleven “action lines.” At the same time 

 in the same venue, the inaugural UN Global Dialogue on AI (UNGDAI) will take place, convening government officials, company representatives, and civil society organizations — 

 heads of state and company leaders — “to discuss international cooperation, share best practices and lessons learned, and to facilitate open, transparent and inclusive discussions on artificial intelligence governance.”

This overlap is intentional. AI governance initiatives have proliferated locally and globally over the last two years, and the UNGDAI has the potential to help build coherence across those efforts. While that is a daunting task, the WSIS process offers valuable lessons on how (and how not) to attempt such governance coordination. Perhaps more importantly, it can help illuminate the technical and social arrangements that govern the Internet, which constitutes the technical and social substrate upon which AI is built.

, and learning lessons should not be equated with rote repetition or modeling. That said, too many of the leading stakeholders participating in AI governance seem unfamiliar with — or uninterested in — this deeply intertwined and relevant history. The overlap between WSIS and the UNGDAI presents an important learning opportunity.

Last year, the WSIS+20 review process offered an opportunity to reflect on the relevance of the WSIS framework, understand its role within the evolving digital governance landscape, and reimagine what a people-centered, inclusive, and development-focused information society means today. Throughout that review, key questions included how emerging UN processes like the UNGDAI can and should coexist with WSIS, how they can complement and strengthen each other, and what lessons from WSIS can be applied to these new processes. When answering these questions, it is helpful to analyze what WSIS did well, what it was wise to avoid, and what it did not do well.

. When WSIS was initially convened, the Internet was just beginning to scale and policymaking on relevant topics such as spectrum allocation, infrastructure buildout, and liability rules was largely defined by national governments. But WSIS, reflecting the wide spectrum of organizations involved with and interested in the early Internet, convened an unprecedented range of civil society organizations, technologists, companies, youth organizations, and religious leaders. WSIS was the first global process to formally embrace a multistakeholder approach to governing digital technology, recognizing that “governments, as well as the private sector, civil society and … international organizations have an important role and responsibility in the development of the Information Society and, as appropriate, in decision-making processes.”

The first global AI safety summits – in 

 in 2024 – were relatively small convenings, with high-level representatives from OECD governments and tech companies conspicuously overrepresented. The 2025 Paris AI Action Summit featured broader civil society participation, but attendees remained primarily American and European. In February, the 

While next week’s UNGDAI will be a substantially smaller affair, it should feature a range of speakers who can bring diverse perspectives from different stakeholder groups, geographies, experiences, and expertise.

Going forward, it will be critical for the UNGDAI’s secretariat and the co-chairs of the second UNGDAI, which will take place in New York in May 2027, to think creatively about how to ensure broad and diverse representation, especially from marginalized communities. In doing so, they should draw on 

 forged over decades of Internet governance experience and recent examples such as the 

Informal Multistakeholder Sounding Board,

 which was established to drive diverse participation in the WSIS 20-year review process.

Elevate Global South priorities and participation

. While WSIS has not been immune from overarching power dynamics and practical limitations that consistently challenge the democratic legitimacy of multilateral processes, it has nevertheless enabled Global South engagement by explicitly linking digital policy to development, including the UN’s 

. National and regional implementation of the WSIS Action Lines enables the framework to be adapted to local realities and needs, and encourages participation from local stakeholders.

It will be critical for the UNGDAI to reflect Global South priorities in its agenda and outcomes. In addition to engaging directly with external initiatives committed to elevating Global South and civil society participation in AI governance, such as 

, it can also align with and support relevant WSIS themes, action lines, activities, and messages.

This can include developing capacity-building mechanisms for AI, encouraging linguistic diversity in LLMs, promoting financing mechanisms to expand access, preserving cultural heritage in digital content, understanding the upstream social and environmental impacts of AI, fostering risk mitigation and redress measures, ensuring that human rights are centered in AI design and deployment, in line with the UN Guiding Principles on Business and Human Rights, and establishing safeguards to prevent adverse impacts arising from the use of digital and emerging technologies – ideas that are reflected in the 

 and have been echoed in UNGDAI consultation 

Articulate a comprehensive approach to access.

 WSIS recognized that meaningful access to digital technologies requires a comprehensive approach—addressing affordability, creative financing, digital literacy, and localized content through relevant policy frameworks. WSIS formalized this approach in the 

, breaking access down into the dimensions of infrastructure, affordability, capacity building, and content.

 on four thematic pillars: (i) AI opportunities and implications; (ii) bridging AI divides; (iii) safe, secure, and trustworthy AI; and (iv) respecting, protecting, and promoting human rights. As these are unpacked, participants would do well to draw connections between these themes and relevant WSIS action lines, as well as to acknowledge that efforts to “bridge” digital divides must be front-and-center and approached holistically and in a multidimensional way.

WSIS avoided rigid standardization by designing a framework based on flexibility, decentralized cooperation, and high-level principles rather than top-down rules. The Geneva Declaration recognized that digital policy needs to be context-specific by encouraging national governments and regional bodies to implement the WSIS Action Lines and by avoiding attempts to consolidate standard-setting authority within UN bodies.

AI standards processes are already proliferating across a wide range of standards bodies, most of which have also been involved in setting and stewarding key Internet protocols and standards. While the UN can play an important role in helping to draw attention to and encourage participation in these processes, it should continue to resist any impulse to govern them.

: WSIS was successful in part because it did not decide to pursue a global treaty or otherwise establish processes to centralize or unify Internet rule-making. Instead, it created an inclusive and adaptable ecosystem for digital policy and development, embracing a multistakeholder approach that acknowledged and created space for discussions of, but not top-down decisions about, the respective roles of governments, the private sector, civil society, and the technical community.

Central to this approach was the establishment of the Internet Governance Forum (IGF), which has become a key global platform for open, inclusive, and non-binding discussions of internet policy. The value of this approach was recognized last year when the WSIS+20 review permanently renewed the IGF’s mandate. The IGF's components, including its network of 

intersessional policy networks and dynamic coalitions

, are vital to maintaining its inclusive, bottom-up nature. The UNGDAI and other AI-focused governance processes would be wise to plug into and build directly upon this inclusive, transparent, local-to-global infrastructure of convenings, expertise, and experience, rather than to try to create something new from whole cloth.

 for its lack of tangible outcomes from dialogue, as well as for 

 to develop effective mechanisms to mitigate digital divides and govern what is now a highly commercialized, geopolitically contested, and rapidly evolving digital ecosystem. While important issues have been identified and discussed within the WSIS framework, and some responsive actions have resulted, its underlying objectives have not been effectively resolved. Currently, there is an active debate on how to make the IGF more effective, with the WSIS+20 outcome document calling for the creation of a government track, enhanced intersessional work, and a stronger Secretariat (this will be the 

These are challenges that the UNGDAI should acknowledge from the jump. While the Scientific Panel can provide timely insights into emerging risks and challenges, and the Dialogue can bring diverse stakeholders together to share knowledge and approaches on urgent issues, the success of the process will depend not just on how inclusive and responsive the actual Dialogues are, but also on how it is able to bring coherence and continuity to the efforts that are organized in between and around them.

The UNGDAI should focus on outcomes, ensuring that its convenings generate relevant and ambitious commitments from all sectors, establishing or embracing existing mechanisms to transparently track progress and drive accountability. The impacts of AI are generating significant attention and resources, which should be leveraged to support this architecture of action, including support for civil society efforts to hold government and private sector actors accountable.

What Needs to Happen in Geneva and Beyond

The co-chairs of the UNGDAI, with support from a Secretariat drawn from relevant UN agencies and offices, have done well to set up an open, inclusive process to help shape the Dialogue. They have 

 closed consultations with member states, as well as public consultations open to all stakeholders; 

 written inputs on process and substance; 

 four thematic clusters around which the agenda will be shaped and 

 mechanisms that promote collaboration. For a UN process that was authorized only 10-months ago, this is an impressive set of steps — especially in light of the broader political and funding challenges the UN system is facing.

Notwithstanding these efforts, a lot of uncertainty remains with less than one week to go. While a skeleton agenda was published last month, key questions remain as to how the UNGDAI will balance its mandates to feature “a multi-stakeholder plenary meeting with a high-level governmental segment,” “a presentation of the [inaugural] annual report of the [Independent International] Scientific Panel,” and other “thematic discussions.” Organizations that 

 “side-event sessions” have only recently heard back about which will be included in the agenda, making travel and logistical coordination challenging.

Latest

Podcast

Recent

Our content delivered to your inbox

Thank you!

Around the globe