AI

Hash matching can slow the spread of NCII, but it is not a silver bullet. It only catches known content, struggles with edits and AI variants, & cannot determine consent or context on its own. That is why governance, not just tech, matters. Read CDT’s brief:

Watching Sam Altman's World ID announcement on its new ID verification partnerships w/ Tinder, Zoom, AWS, etc & want to point out that @technologyreview.com predicted all of this 4 years ago—while pointing out the privacy violations they built on. www.technologyreview.com/2022/04/06/1...

What role should AI play in a healthier digital ecosystem and who gets to decide? In the latest Tech Talks episode, CDT’s Alexandra Reeve Givens + Mozilla Foundation’s Mark Surman dig into the choices shaping our online world. Listen now:

The money is flowing. Companies and investors put $6.1 billion into humanoid robots in 2025 alone, four times what was invested in 2024. What happened? A revolution in how machines have learned to interact with the world.

“What AI changes is scale, speed, and accessibility. The technical barrier has dropped significantly, which means more people...can produce more convincing outputs with minimal effort. As with many AI-enabled harms, this results in a glut of content.” www.wired.com/story/deepfa...

Our Data Center Policy in Practice training series starts on April 22nd at 1 pm ET with “The AI Industry, Data Center Buildout, & How to Take Power Back.” Explore policy interventions and organizing interventions to tackle corporate power. RSVP: bit.ly/policy-in-practice-training-series

“In the last school year across the county, 15 percent of students reported seeing sexually explicit deepfakes of someone associated with their school, according to a recent report by the Center for Democracy and Technology.”

“The true scale of deepfake sexual abuse taking place in schools is likely much higher…[] 15 percent of students surveyed by the Center for Democracy and Technology said they knew about AI-generated deepfakes linked to their school.”

The immediate danger is not that machines will act without human oversight; it is that human overseers have no idea what the machines are actually “thinking.” The Pentagon’s guidelines are fundamentally flawed because they rest on the dangerous assumption that humans understand how AI systems work.

“But today people actually report being in relationships with AI companions. According to a 2025 survey by CDT, about 1 in 5 high school students say they or someone they know has had a romantic relationship with an AI.”

Scammers are buying tools advertised on Telegram that trick banks' facial recognition checks. The stakes are enormous with crypto scams tallying an estimated $17 billion in 2025 alone. Banks and regulators are struggling to keep up.

“People should be able to move through their daily lives without fear that stalkers, scammers, abusers, federal agents, and activists across the political spectrum are silently and invisibly verifying their identities," as face recognition on smart glasses would enable. www.wired.com/story/meta-r...

I am at #CHI2026 representing @datasociety.bsky.social! I'm participating in the Social & Emotional Uses of AI workshop, where I'll be contributing findings from our longitudinal study on how people use LLMs for mental & emotional care. If you're around, please say hi! @chi.acm.org

In an effort to reduce labor costs, companies pushing genAI are are instead yielding a deluge of "workslop." GenAI is often presented "as a general-use tool that can do anything, but the reality doesn’t work that way,” says D&S's @aihathing.bsky.social. www.theguardian.com/technology/2...

CDT’s Travis Hall in the The Washington Post: “If you look at the legislation, it actually does regulate the way in which AI is being used for consequential decisions like people’s health care and their employment.” www.washingtonpost.c...

At "the racing heart of the modern AI horror genre" is a fear not of intelligence but of desire, Amanda Gefter writes: "A machine that knows a lot doesn’t scare us. A machine that *wants* something does." www.quantamagazine.org/why-do-we-te...

We have a new list of noteworthy AI trends and advances—and we’re excited to share it with you.

“Crawford of CDT says HIPAA isn’t failsafe as it protects users only from the institutions that process and store data, not the data itself. Which means AI companies, not coming under the ambit of such law, have no reason to worry while stealing it or forming users to deliver it.”

. @heidykhlaaf.bsky.social called for “independent, rigorous evaluations that objectively assess the risk of tools within existing cyber laws. Democratic institutions should determine how these models are released and deployed, not a private corporation with its own incentives." shorturl.at/4dkAm

“A lot of the ways that AI systems play out now are just a reiteration of different kinds of algorithmic controls and disempowerment of workers that we’ve already seen before,” says Labor Futures Researcher (and Siegel Research Fellow) @cariatida.bsky.social. www.siegelendowment.org/insights/ale...

In an industry that doesn’t stand still, Stanford’s AI Index, an annual roundup of key results and trends, is a chance to take a breath. (It’s a marathon, not a sprint, after all.)

Today! Join the Mystery AI Hype Theater 3000 livestream to hear @emilymbender.bsky.social and @alexhanna.bsky.social chat with very special guest Carmen Maria Machado noon PT, twitch.tv/dair_institute

Remember the cyberattacks on the U.S. government in the early 2020s? The Biden admin asked tech companies to help and got some “free” upgrades. The Trump admin is doing similarly with AI. But our reporting shows there’s no such thing as a free lunch.

And when opposing all regulation seems untenable? Companies place themselves in the driver’s seat by laying out their own alternative platforms with all the right sounds but no teeth: www.wsj.com/tech/ai/ai-c...

We spoke with @wsj.com and @theguardian.com this week on how to make sense of OpenAI's latest policy rebrand. While the doc has a lot of the sounds of wanting more regulatory oversight, the company has lobbied aggressively for a deregulatory stance toward AI: www.theguardian.com/technology/2...

CDT’s Anex-Ries: “It’s not only heightening the risk of failed AI projects and wasted taxpayer dollars, but it also opens the American public to a host of potential AI harms.”

CDT’s Quinn Anex-Ries: “Without these kinds of measures in place, some of the riskiest tools used by federal agencies are left without real oversight or mechanisms for agencies to validate performance and efficacy.”

An analysis of Google’s AI Overviews found that they were accurate approximately 9 out of 10 times. That might sound pretty good, but given the trillions of searches Google processes, this means it provides “tens of millions of erroneous answers every hour.” www.nytimes.com/2026/04/07/t...

If you’re following AI news, you’re probably getting whiplash. AI is a gold rush. AI is a bubble. AI is taking your job. AI can’t even read a clock. The 2026 AI Index, AI’s annual report card, comes out today and cuts through some of that noise.

On the next Mystery AI Hype Theater 3000: @emilymbender.bsky.social and @alexhanna.bsky.social will be speaking to author Carmen Maria Machado about the use of LLMs in writing, from language arts classes to the publishing industry. Listen live: Monday, April 13, noon PT, twitch.tv/dair_institute

For more, see our work: ainowinstitute.org/publications...

On @channel4news.bsky.social @heidykhlaaf.bsky.social explains the stakes of Anthropic’s Mythos and its impact on security. We lack the evidence and independent expertise needed to verify the claims around the security tool - especially problematic given the risks posed by LLMs. shorturl.at/kNtU9

"Glasswing is built on a deeply uncomfortable premise — that the only way to protect us from dangerous AI models is to build them first," @caseynewton.bsky.social writes. "And Anthropic is doing so in an environment that is barely regulated at all." www.platformer.news/anthropic-my...

Upwards of 12 percent of the population struggles with health anxiety, and turning to AI chatbots for health advice can make it worse. www.theatlantic.com/technology/2...

Two months before Calvin Alexander’s parole hearing date, he was told he was no longer eligible. Why? An algorithm had deemed the nearly blind 70-year-old, who uses a wheelchair, a “moderate risk,” barring him from speaking to the parole board. (Published April 2025 w/ @veritenews.org)

CDT urges NIST to adopt standards that incorporate civil rights principles, incl. anti-discrimination measures like disparate treatment + disparate impact testing, & encouraging the development of domain-specific standards like housing, lending, education, & employment.

CDT joined a civil society letter to the National Institute for Standards and Technology (NIST) responding to the agency’s draft guidance on practices for automated benchmark evaluations of language models.

The results of a new Gallup survey suggest that Americans’ animosity toward AI extends to a younger generation — one that is currently struggling to find its footing in the workplace. www.nytimes.com/2026/04/09/s...

The Pentagon’s race to adopt AI without effective regulation poses serious — even deadly — risks. Congress must establish guardrails. Learn more in our report on the military and AI: bit.ly/4sJCA2S

“Transparency should be a ‘foundational piece of any effort’ on a national AI framework and shouldn’t be framed as an ‘additional burden that’s being imposed,’ Bogen said. ‘We actually need this if people want this technology to go anywhere.’”

“Major AI developers [] have moved to adopt their own safety frameworks & publicly document some info ab their models to address concerns. The bill would help clarify, offer guidance, & standardize that process at the federal level, said [CDT’s] Miranda Bogen

“In fact, according to a 2025 report from CDT, which included surveys from students in grades 9-12, only 22% of students said they received guidance on their school’s AI policy, yet 86% of students say they have used AI during the last school year.”

The Pentagon’s race to adopt AI without effective regulation poses serious — even deadly — risks. Congress must establish guardrails. Learn more in our report on the military and AI: bit.ly/4sJCA2S

As companies like OpenAI test ads, AI systems could become more personalized and more persuasive, while raising risks around trust, manipulation, and profit-driven outputs. “Watch the business models.” Read more:

In The Globe and Mail, CDT’s Miranda Bogen argues that ads coming to AI chats aren’t surprising — they’re a warning sign. We’ve seen this playbook before with social media, and the incentives are hard to resist.

The CEO of Microsoft AI writes about the complex and fast-moving reality powering AI development.

🔇🔇🔇 On Friday April 17th (9am PDT/5pm CET/7pm EAT), we'll have a virtual event introducing our Refugees, Migrants and AI research program which centers the lived experiences of refugees and migrants and puts power back into their hands🧵 ☘️☘️☘️ Register at www.eventbrite.com/e/refugees-a...

The Pentagon’s race to adopt AI without effective regulation poses serious — even deadly — risks. Congress must establish guardrails. Learn more in our report on the military and AI: bit.ly/4sJCA2S

Join @emilymbender.bsky.social and I in New York for Mystery AI Hype Theater 3000’s first LIVE show! We’re doing a live taping at the Starr Bar in Brooklyn on Thursday, April 30. Space is limited, so get your tickets. All profit goes to @dairinstitute.bsky.social. www.eventbrite.com/e/mystery-ai...

Read the research! "[G]enerative AI technologies make it far easier for companies to radically manipulate models’ images and repurpose them,” write @cariatida.bsky.social, Zoë West, and Sanjay Pinto. datasociety.net/points/fashi...

"I want to stay provocative in our thinking — the more we talk about [AI] and the more we can think about it, we won't get so controlled or seduced by it," says Pamela Anderson, in an article that cites D&S research on how AI is reshaping modeling work. www.marieclaire.com/fashion/cele...

“Two years ago, there was an investigation showing that Maven can have 30% or less accuracy, which isn't far from indiscriminate targeting.” @heidykhlaaf.bsky.social joins @telegraph.co.uk to unpack the risks of systems like Maven and what safeguards could look like: youtube.com/watch?v=NdHYDGHN5rQ

Exactly! The underwater data centers of the 2010s weren't conceived because computers needed cooling but because it was a fun idea, like underwater restaurants.

CDT’s Maddy Dwyer: “Through the [GovAI] coalition’s network, gov, civil society and industry leaders can share knowledge & discuss ideas, pushing the public sector towards responsible and trustworthy AI adoption and use in the delivery of public services.”

Proposed rules about how the government buys and deploys AI could allow bureaucrats to sidestep necessary safeguards and supercharge mass surveillance. www.eff.org/deeplinks/2...

Business owners and e-commerce experts told MIT Technology Review that AI tools are making sourcing more accessible and significantly shortening the time it takes to go from product idea to launch.

The very existence of the "constitution" Anthropic created for its LLM, Claude, "is a product of the larger crisis of constitutionalism in the United States,” Jill Lepore writes. www.newyorker.com/magazine/202...

NEW: We’ve been reporting on cybersecurity for years. As President Donald Trump and his Cabinet say artificial intelligence will transform the nation, the messaging isn’t new. It follows a familiar pattern.

1. Read “Empire of AI” by @karenhao.bsky.social 2. Watch this talk between Karen and @timnitgebru.blacksky.app 3. Follow @dairinstitute.bsky.social, Karen and Timnit for a sober, critical analysis of the “AI” field.

The incentives need to be on humans not to publish things that are p hacked or not generalizable. Human beings will then use AI to ensure that their name is not attached to shoddy work. We are in charge of AI, we can make it do whatever we want. If it doesn't do what we want we can train and repair.

There must also be incentives. Penalties for having some things p-hacked or unrobust with your name on it. AI can help you cheat,and AI can help you do a better and more generalizable science. It's about incentives, what we're rewarding, and what the review process is promoting.

Once things are engineering problems, they see progressive improvement. AI on the reviewing side can test robustness under alternative specifications, evidence of p hacking, etc. AI can exert effort human reviewers won't, extending their abilities. The rest is incentives.

I think AI is more likely to tell people they’re misusing stats than to get people to misuse them. People have generally been proceeding with little understanding and the only way is up.

Proposed rules about how the government buys and deploys AI could allow bureaucrats to sidestep necessary safeguards and supercharge mass surveillance. www.eff.org/deeplinks/2...

CDT’s Bogen: “Who is keeping track of things about users? And how is that information being used to make determinations about them?”

"[Bogen says] AI companies are reprising a similar privacy problem: users hand these systems enormous amounts of info precisely because the tools are designed to know them well enough to be useful.”

“[CDT’s] Bogen says the pattern is familiar from older Internet platforms, where small behavioral cues became signals that shaped what users saw & how they were categorized.

CDT’s Bogen: “Even if it’s a very legible and very simple prediction pattern, how you use that information is a separate governance question.”

“A signal collected for one purpose can migrate into other parts of a product in ways users neither expect nor consent to.”

“Miranda Bogen, director of the AI Governance Lab at the CDT, says the more pressing issue is what happens to such information once a company has it.”

Specifically, we address how requirements for AI systems to produce outputs regardless of potential risks and for vendors to adhere to meaningless “unbiased AI principles” are overly broad and would make AI tools less safe.

CDT comments highlight how the draft AI Terms and Conditions could undermine responsible AI practices in federal procurement.

Today, CDT, the @eff.org, the Protect Democracy, & EPIC submitted comments to the GSA in response to the request for comment on the draft AI Terms and Conditions for federal solicitations and contracts. cdt.org/insights/cen...

Days into the new baseball season, @jasonkoebler.bsky.social looks at how the early disputes driven by Major League Baseball’s new automated ball-strike challenge system might be seen as “a microcosm of various human-AI beefs in society more broadly." www.404media.co/you-cant-def...

A former DOGE staffer developed an AI tool that hallucinated the size of Veterans Affairs contracts. For example, it concluded that more than a thousand deals were each worth $34 million, when in fact some were for as little as $35,000. (Published June 2025)

CDT’s Anex-Ries: “Implementing GSA's draft terms and conditions across all AI solicitations and contracts in the federal government could result in systems with fewer safety protections and worse outcomes, and a vendor community that is reticent to push back when their products are used unsafely.”

“[CDT’s] Anex-Ries said these latest updates to previous GSA memos on AI acquisition may offer clarity but include terms that would undermine ‘key safety measures in AI systems.’”

“Quinn Anex-Ries, a senior policy analyst at the CDT, said the draft terms & conditions serve as the most recent resource to help guide companies and AI developers in working with the govt, & that it centers on ‘unbiased AI principles’ for large language models.”

When a research participant referred to his AI companion as a “diet girlfriend,” Research Assistant @merylye.bsky.social saw a useful metaphor: chatbots lack social nutrition; they “offer comfort in the moment, but are rarely filling on their own." datasociety.net/points/socia...

“’Human in the loop’ is necessary, but the process itself makes the loop stultifying, and encourages the human to take themselves out of the loop. That process is straight up dangerous. The temptation to let it rip is always there, and I didn't even have a boss pressuring.“ #genai+writing

AI is cause for consternation.⛅️ But it’s not all doom & gloom.☀️ Humans aren't the #DataOil. We're the Kami-supported superintelligence tilling the #DataSoil!🧑‍🌾 Join The Human Movement & Let’s #FreeTheFuture🖖 — together. ▶️ human.mov ▶️ civic.ai

look, i am an artist and tech geek on the African continent so I do not include any genAI or "general purpose ai" (lol). we always get the shorter end of the stick here. i do believe that low resource Machine Learning will be needed. See @dairinstitute.bsky.social www.dair-institute.org/blog/

WATCH: CDT’s Jake Laperruque on NBC News on the negative impacts of facial recognition technology, in response to the latest reports that a woman was wrongfully arrested after an AI facial recognition tool falsely linked her to bank fraud:

NEW: Mystery AI Hype Theater 3000 - Data Centers Go Nuclear In which @mwoluchem.bsky.social and @livgar.bsky.social join @emilymbender.bsky.social and @alexhanna.bsky.social to get into data center infrastructure build out and resistance in Pennsylvania. www.buzzsprout.com/2126417/epis...

Explore the North Star Data Center Policy Toolkit and register for the training series: datacenters.ainowinstitute.org bit.ly/policy-in-pr...

Created with community organizations, advocates, and policymakers, the toolkit offers a North Star, transformative approach and practical roadmap to achieve that vision, covering interventions from bans and moratoriums to zoning, water, energy, transparency, air quality, and beyond.

Hundreds of communities are being asked to absorb the expansion of AI data centers, despite their harm to our scarce natural resources, public budgets, health, and futures. Many existing ordinances and proposed regulations stop short of transformative policy.

Our accompanying Data Center Policy in Practice training series starts April 22. Co-hosted with the Data Center Working Group, each session features expert-led policy analysis and real-world strategies from local and state organizers: bit.ly/policy-in-practice-training-series

Today, we're launching our interactive North Star Data Center Policy Toolkit to stop, slow, and restrict the rapid expansion of US AI data centers. The toolkit and training series lay out affirmative local, state, and federal policies to protect communities: datacenters.ainowinstitute.org 🧵

People in Nigeria and India are strapping iPhones onto their heads and recording themselves doing chores.

Given how little access to healthcare many people have, health chatbots could have real benefits — but the lack of independent testing makes it impossible to know whether those benefits outweigh the risks. www.technologyreview.com/2026/03/30/1...

One-off tests don’t measure AI’s true impact. We’re better off shifting to more human-centered, context-specific methods.

Talking to some undergrads tomorrow about the why no one actually pays us to type model.fit() and asked them ahead of time to submit answers to a "trivially easy" data problem Fun to see the results come it, wild to know how many business run with a not dissimilar fog of war

Tonight’s the night! 🎉 Join CDT’s Spring Fling for an evening of cocktails, connection, and conversation with leaders in privacy and AI. No ticket yet? No problem—tickets will be available at the door. See you there! 🥂

The military is rolling out an “AI-first” approach to warfare. But the recent dispute between the Pentagon and AI firm Anthropic raises questions about whether the military’s deployment of the technology is effective, safe, and lawful. bit.ly/40rNSg4

Without guardrails, like transparency, audit trails, and limits on use, AI risks obscuring accountability in policing. Officers’ judgment can’t be outsourced. Civil rights must come before convenience.

New from CDT’s Tom Bowman: Automated police report drafting tools promise efficiency—but at what cost? AI-generated narratives can shape arrests, charges, and sentencing. When errors or bias slip in, the consequences can impact someone’s liberty.

Decisions to tweet first and lawyer later didn’t sit well with a federal judge, who last week halted the government’s punishment of the AI company.

Specialized chatbots might make a difference for people with limited health-care access. Without more testing, we don't know if they’ll help or harm.

"The tech sector’s sudden emphasis on robotics is fueling concerns about whether it could wipe out vast swaths of an already-devastated manufacturing sector, accelerating declines over the last half-century as such work has been outsourced overseas." www.washingtonpost.com/technology/2...