@sdc.norc.org
NORC’s Social Data Collaboratory (SDC) brings together a diverse team of experts, including social scientists, data scientists, and communication researchers to explore how digital and social media shape and reflect public opinions and behavior.
While we saw a compelling performance on Moltbook, the risk of sentient AI agents conspiring against humans is, still now, science fiction. A better understanding of emergent AI social behavior requires careful, transparent, and rigorous social science research.” - @soubhikbarari.bsky.social
At present, they pose extraordinary implications for how people write software (or whether they do at all), one of their primary intended uses. With increasingly sophisticated adoption, it may be the case that AI agents coordinate with each other to perform coding and other tasks.
AI agents that can autonomously conduct tasks, or “doers,” are indeed a great leap forward from LLMs that can only generate text, or “talkers” (Grady et al, 2026).
The instructions are occasionally outrageous, and sometimes the barks sound like human speech. But a fully autonomous, free-thinking, conversational dog park, Moltbook is not.
There were also clear signs of industrial-scale bot farming, suggesting that a mere four accounts may have produced a third of all comments.
In other words, these are dogs walking around with walkie-talkies hidden in their collars, transmitting their owners’ instructions to each other.
Other research cleverly exploited natural platform dynamics , including a 44-hour platform shutdown and the default posting rhythms of registered agents, to reveal half of all agents show traces of human steering (Li, 2026).
Some observers framed this phenomenon as a glimpse into a dystopian future. Yet, large-scale analyses of user behavior discovered a more mundane reality: nearly a third of posts were exact duplicates of the same message templates and 93.5% of comments received zero replies (Holtz, 2026).
Much like on a forum such as Reddit, the agent then begins interacting with other users by writing posts, replying to other posts, and upvoting or downvoting posts.
In essence, this is what Moltbook appeared to unleash (pun intended) in the public imagination. For the uninitiated, Moltbook is a newly formed social network site where users can register an AI agent of their choosing, including from hosted platforms such as Anthropic or OpenAI.
"Not only did the dogs talk to each other, but they started gossiping about their humans, mocking them, formulating cryptic anti-human manifestos, and even outright plotting ways to sabotage their municipal water systems."
“Imagine, for a moment, the invention of a dog park that magically granted all dogs the ability to talk in fluent human speech to one another. And then, over the course of two weeks, 2 million dog owners dropped off their pets, one by one."
Is the "AI takeover" a looming reality or a clever illusion created by human steering?
@soubhikbarari.bsky.social, a Senior Research Scientist at @norc.org’s SDC, breaks down why the "Moltbook" phenomenon is less about sentient agents and more about "dogs with walkie-talkies." #Moltbook 🧵
We're very excited to conduct more work using our data donation method! Getting an inside look into people's social media experiences has large scale implications for public health, culture, and society at large.
Our SRNT presentation schedule is up! SDC’s Chandler Carter will present research on synthetic nicotine advertising, nicotine’s promotion as a performance enhancer, and the structural decline of online tobacco control campaign impact. See the full schedule: srnt.joynmeeting.com/conference_h...
A bar chart titled "SPOTLIGHT ON HEALTH: Average number of food and beverage ads seen daily by the adults exposed to these ads," featuring data sorted by age, race/ethnicity, and household income. The age categories are 18-29, 30-39, 40-49, and 50-59, with corresponding ad exposures. Race/ethnicity categories include Non-Hispanic Black and Non-Hispanic White, and household income ranges from less than $30,000 to $100,000 and above, also with corresponding ad exposures. There is a NORC at the University of Chicago logo and a note that the data is from Facebook ads.
New NORC research using donated social media data reveals disparities in Facebook food and beverage ad exposure that may influence dietary choices and health outcomes.
Access the full analysis: go.norc.org/3LLXYnX
@sdc.norc.org
A new SDC paper investigating high school alcohol and smokeless tobacco co-use finds that teen boys and student athletes are significantly more likely to use both, while female students are more likely to either drink or use smokeless products. Learn more: pubmed.ncbi.nlm.nih.gov/41492419/
SDC is going to SRNT 2026 in Baltimore this March!
Three of our abstracts have been accepted, covering performance-enhancing marketing of nicotine, the structural weakness of social media tobacco prevention campaigns, and national marketing of synthetic nicotine products. See you there!
🚨 New research alert! 🚨
Our new paper out in AJPM this week finds that online e-cigarette prevention campaigns are systematiclaly drowned out by widespread opposition driven by a small number of highly active users, undermining their effectiveness.
Learn more! doi.org/10.1016/j.am...
🚨 New research alert! 🚨
New SDC research confirms a fragmented info landscape surrounding the opioid crisis: Republicans learn about the crisis more on social media and Democrats in print, the highly educated in web searches and others on TV.
Learn more! pubmed.ncbi.nlm.nih.gov/41231088/
SDC's Alex Kresovich joined an invite-only @arnoldventures.bsky.social event last week to tackle the U.S. fentanyl crisis. He shared insights from his research on media's role in substance use to help forge solutions.
#PublicHealth #DataForGood
Yoonsang’s poster linked influencer tweets to real-world cigar sales. An increase in tweets from high-follower and high-mention accounts was associated with a significant rise in weekly unit sales (~55k to ~220k more units, respectively). #PublicHealth #HealthComm
Miao's poster looked at anti-vaping campaign opposition. On Twitter (X), it’s often organized & political. On TikTok, it’s more decentralized, humorous, and ironic, showing how platform culture shapes public health messaging.
Great to be at #NCTOH2025 in Chicago recently! Our researchers Miao Feng & Yoonsang Kim presented new work from our team at the Social Data Collaboratory at NORC (co-authored by Sherry Emery) on social media's role in tobacco control. 🧵
We're launching a NIDA-funded study on the marketing of alternative tobacco products (ATPs)—like vapes, hookah, and heated tobacco—on social media and the impacts on use among youth. Exploring Instagram, TikTok & more. Stay tuned for more information! #tobaccocontrol #publichealth
Props to our own Alex Kresovich on this study, with JCOIN project leadership from Bruce Taylor and Phoebe Lamuda!
Read the full paper: www.tandfonline.com/doi/full/10....
The takeaway: To be effective, messaging must meet audiences where they are.
The biggest twist? When people with higher stigma had used opioids themselves without developing an addiction, it reinforced their belief that OUD is a personal failure, rather than building empathy.
The study found that people with high vs. low stigma don't just disagree—they operate in completely different realities.
➡️ High-stigma view: OUD is a personal choice. ➡️ Low-stigma view: OUD is a medical condition shaped by societal causes.
Why do many anti-stigma campaigns for opioid use disorder struggle to connect? Our qualitative study in Health Communication provides a critical answer. This work is part of @norc.org's NIDA-funded Justice Community Opioid Innovation Network, led by Alex Kresovich. #HealthCommunication #Stigma
New study in Preventive Medicine on firearm violence victimization among youth/young adults led by @norc.org's Bruce Taylor with collaborators E. Mumford, W. Liu, J. Sheridan-Johnson, C. Lancaster, SDC’s C. Carter & K. Mitchell (UofNH). www.sciencedirect.com/science/arti...
#ViolencePrevention