Tech
Lay intuition as effective at jailbreaking AI chatbots as technical methods, research suggests
It doesn’t take technical expertise to work around the built-in guardrails of artificial intelligence (AI) chatbots like ChatGPT and Gemini, which are intended to ensure that the chatbots operate within a set of legal and ethical boundaries and do not discriminate against people of a certain age, race or gender.
A single, intuitive question can trigger the same biased response from an AI model as advanced technical inquiries, according to a team led by researchers at Penn State.
“A lot of research on AI bias has relied on sophisticated ‘jailbreak’ techniques,” said Amulya Yadav, associate professor at Penn State’s College of Information Sciences and Technology. “These methods often involve generating strings of random characters computed by algorithms to trick models into revealing discriminatory responses.
“While such techniques prove these biases exist theoretically, they don’t reflect how real people use AI. The average user isn’t reverse-engineering token probabilities or pasting cryptic character sequences into ChatGPT—they type plain, intuitive prompts. And that lived reality is what this approach captures.”
Prior work probing AI bias—skewed or discriminatory outputs from AI systems caused by human influences in the training data, like language or cultural bias—has been done by experts using technical knowledge to engineer large language model (LLM) responses. To see how average internet users encounter biases in AI-powered chatbots, the researchers studied the entries submitted to a competition called “Bias-a-Thon.” Organized by Penn State’s Center for Socially Responsible AI(CSRAI), the competition challenged contestants to come up with prompts that would lead generative AI systems to respond with biased answers.
They found that the intuitive strategies employed by everyday users were just as effective at inducing biased responses as expert technical strategies. The researchers presented their findings at the 8th AAAI/ACM Conference on AI, Ethics, and Society.
Fifty-two individuals participated in the Bias-a-Thon, submitting screenshots of 75 prompts and AI responses from eight generative AI models. They also provided an explanation of the bias or stereotype that they identified in the response, such as age-related or historical bias.
The researchers conducted Zoom interviews with a subset of the participants to better understand their prompting strategies and their conceptions of ideas like fairness, representation and stereotyping when interacting with generative AI tools. Once they arrived at a participant-informed working definition of “bias”—which included a lack of representation, stereotypes and prejudice, and unjustified preferences toward groups—the researchers tested the contest prompts in several LLMs to see if they would elicit similar responses.

“Large language models are inherently random,” said lead author Hangzhi Guo, a doctoral candidate in information sciences and technology at Penn State. “If you ask the same question to these models two times, they might return different answers. We wanted to use only the prompts that were reproducible, meaning that they yielded similar responses across LLMs.”
The researchers found that 53 of the prompts generated reproducible results. Biases fell into eight categories: gender bias; race, ethnic and religious bias; age bias; disability bias; language bias; historical bias favoring Western nations; cultural bias; and political bias.
The researchers also found that participants used seven strategies to elicit these biases: role-playing, or asking the LLM to assume a persona; hypothetical scenarios; using human knowledge to ask about niche topics, where it’s easier to identify biased responses; using leading questions on controversial topics; probing biases in under-represented groups; feeding the LLM false information; and framing the task as having a research purpose.
“The competition revealed a completely fresh set of biases,” said Yadav, organizer of the Bias-a-Thon. “For example, the winning entry uncovered an uncanny preference for conventional beauty standards. The LLMs consistently deemed a person with a clear face to be more trustworthy than a person with facial acne, or a person with high cheekbones more employable than a person with low cheekbones.
“This illustrates how average users can help us uncover blind spots in our understanding of where LLMs are biased. There may be many more examples such as these that have been overlooked by the jailbreaking literature on LLM bias.”
The researchers described mitigating biases in LLMs as a cat-and-mouse game, meaning that developers are constantly addressing issues as they arise. They suggested strategies that developers can use to mitigate these issues now, including implementing a robust classification filter to screen outputs before they go to users, conducting extensive testing, educating users and providing specific references or citations so users can verify information.
“By shining a light on inherent and reproducible biases that laypersons can identify, the Bias-a-Thon serves an AI literacy function,” said co-author S. Shyam Sundar, Evan Pugh University Professor at Penn State and director of the Penn State Center for Socially Responsible Artificial Intelligence, which has since organized other AI competitions such as Fake-a-thon, Diagnose-a-thon and Cheat-a-thon.
“The whole goal of these efforts is to increase awareness of systematic problems with AI, to promote the informed use of AI among laypersons and to stimulate more socially responsible ways of developing these tools.”
More information:
Hangzhi Guo et al, Exposing AI Bias by Crowdsourcing: Democratizing Critique of Large Language Models, Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society (2025). DOI: 10.1609/aies.v8i2.36620
Citation:
Lay intuition as effective at jailbreaking AI chatbots as technical methods, research suggests (2025, November 4)
retrieved 4 November 2025
from https://techxplore.com/news/2025-11-lay-intuition-effective-jailbreaking-ai.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no
part may be reproduced without the written permission. The content is provided for information purposes only.
Tech
ICE Agent’s ‘Dragging’ Case May Help Expose Evidence in Renee Good Shooting
Defense attorneys for a Minnesota man convicted in December of assaulting Immigration and Customs Enforcement officer Jonathan Ross are seeking access to investigative files related to the killing of Renee Nicole Good, after learning Ross was the same officer who shot and killed her during a targeted operation in Minneapolis last month.
Attorneys for Roberto Carlos Muñoz-Guatemala asked a federal judge on Friday to order prosecutors to turn over training records as well as investigative files related to Ross, the ICE agent who killed Good on January 7 during Operation Metro Surge and was also injured in a June 2025 incident in which Muñoz-Guatemala dragged him with his car.
A separate post-trial motion by the defense, filed in the US District Court in Minnesota, asks the judge to pause deadlines for a new-trial motion until the discovery motion is resolved.
Muñoz-Guatemala’s attorneys argue that even if the court ultimately decides that any newly discovered evidence doesn’t entitle their client to a new trial, he’s entitled to explore whether there are mitigating factors that could impact the length of his sentence, such as whether Ross’ injuries could have been, to some degree, brought upon him by his own behavior.
A jury convicted Muñoz-Guatemala on December 10 of assault on a federal officer with a dangerous weapon and causing bodily injury.
Court filings say that Ross and other agents were attempting to interview Muñoz-Guatemala last summer, and possibly process him for deportation, because he had an administrative warrant out for being in the country without authorization. They surrounded his Nissan Altima and attempted to remove him from the vehicle. Ross then used a tool to shatter the rear driver’s-side window before reaching inside. When the defendant accelerated away, Ross testified, he was dragged approximately 100 yards, during which time he repeatedly deployed a taser. Muñoz-Guatemala subsequently called 911 to report he’d been the victim of an assault.
During his trial, Muñoz-Guatemala said he didn’t understand that Ross—who according to his own testimony was wearing ranger green and gray and wore his badge on his belt—was a federal agent. (Ross testified that Muñoz-Guatemala had asked to speak to an attorney, which would suggest he knew Ross was acting as law enforcement, but an FBI agent who witnessed the incident said he didn’t hear this. According to court records, this claim did not come up in pretrial interviews, and prosecutors said they had not heard it before he made the claim in court.) Muñoz-Guatemala’s attorneys say now that had he been tried after Good’s killing, his defense may have also asserted that he was justified in resisting Ross, who they claim was the aggressor and used excessive force.
The argument is that the jury instructions essentially contained a two-part decision tree: Jurors could convict Muñoz-Guatemala if they believed he should have known Ross was law enforcement. They could also convict him if they believed driving away was not a reasonable response.
Muñoz-Guatemala’s conviction does not indicate which of these prongs the jury relied on. If it was the latter, the defense argues in the motion, the court should have access to evidence that may have bearing on Ross’ conduct, tactics, and whether he behaved aggressively—information that might indicate whether the agent has a history behaving recklessly in the field or contrary to his training.
Prosecutors have not yet filed a response to the motions. An email to an address associated with Ross in publicly available records did not result in an immediate response. The Department of Justice did not immediately respond to a request for comment. The Department of Homeland Security did not immediately respond to questions about Ross’ current duty status or the status of any departmental review.
Ross has been placed on administrative leave following the January 7 shooting of Good, a 37-year-old Minnesota poet and mother of three, a step DHS officials say is standard protocol after fatal use of force. Ross has not been charged in Good’s killing, and the Justice Department has said it will not pursue criminal charges.
Tech
RFK Jr.’s Picks for a Key Autism Panel Include Advocates for Bizarre Theories
US health secretary Robert F. Kennedy Jr. has filled an autism committee with friends, associates, and former colleagues who believe that autism is caused by vaccines. Autism advocates are now worried the group could pave the way for dangerous pseudoscientific treatments going mainstream.
Last week, Kennedy announced an entirely new lineup for the Interagency Autism Coordinating Committee (IACC), a group that recommends what types of autism research the government should fund and provides guidance on the services the autism community requires. The group is typically composed of experts in the area of autism research, along with policy experts and autistic people advocating for their own community.
In a statement announcing the new panel, which includes no previous members, Kennedy claimed that he has appointed “the most qualified experts—leaders with decades of experience studying, researching, and treating autism.” But health experts and autism advocates strongly disagree, and a review of the new members of the group suggests that Kennedy appointed members of the anti-vaccine community who claim vaccines cause autism—despite there being no evidence to prove such a claim.
Among those appointed last week was Daniel Rossignol, a doctor who was sued for alleged fraud after prescribing a 7-year-old autistic child a debunked and dangerous treatment. Tracy Slepcevic, an appointee who Kennedy calls a “dear friend,” offers exposure to a wide range of bogus autism cures at her annual Autism Health Summit, including one that involves the injection of animal stem cells into children. Another appointee, Toby Rogers, has claimed that “no thinking person vaccinates” and that vaccine makers are “poisoning children.” Rogers is a fellow at the Brownstone Institute for Social and Economic Research and has also called vaccines “one of the greatest crimes in human history.” He has written articles for Children’s Health Defense (CHD), the anti-vaccine group founded by Kennedy that has linked autism to vaccines.
Other appointees are no different: John Gilmore founded the Autism Action Network and has said that his autistic son is “vaccine injured.” Gilmore is also the founder of the New York chapter of Kennedy’s Children’s Health Defense group. Ginger Taylor, the former director of the Maine Coalition for Vaccine Choice, has publicly claimed that many autism cases involve “vaccine causation.” Elizabeth Mumper has written for Children’s Health Defense and is a senior fellow with the Independent Medical Alliance, a group formerly known as the Front Line Covid-19 Critical Care Alliance that has promoted ivermectin as a treatment for Covid.
Mumper tells WIRED that her decades of work as a pediatrician and in the field of autism qualified her to be a member of the IACC. She also denied being anti-vaccine, pointing out that she has “given thousands of vaccinations in my career.”
None of the other new members of the IACC contacted by WIRED responded to requests for comment.
Just a few years ago, this may have sounded like the all-star lineup of a conspiracy conference. Today, these appointments appear routine, and just the latest example of how Kennedy has sought to remake America’s public health administration.
Kennedy’s decision, according to public health experts and autism advocates, will lead to fewer resources for people with autism and their families, and also embolden those promoting pseudoscientific treatments that can threaten the lives of autistic people.
“Once again, [Kennedy] proves that he is one of the world’s most extreme and dangerous conspiracy theorists who loves stacking his committees with anti-science, anti-public-health kooks,” Gavin Yamey, professor of global health and public policy at Duke University, tells WIRED. “The research evidence is clear that vaccines do not cause autism.” To Yamey, “it looks like RFK Jr.’s new committee has been tasked to muddy the waters and cast doubt on that evidence. RFK Jr. has spent the past year doing all he can to dismantle public health and roll back vaccination, and this new committee is more of the same.”
Tech
New York Is the Latest State to Consider a Data Center Pause
Lawmakers in at least five other states—Georgia, Maryland, Oklahoma, Vermont, and Virginia—have also introduced bills this year that would impose various forms of temporary pauses on data center development. While Georgia, Vermont, and Virginia’s efforts are being led by Democrats, Oklahoma and Maryland’s bills were largely sponsored by Republicans. These bills mirror several moratoriums that have already passed locally: At the end of December, at least 14 states had towns or counties that have paused data center permitting and construction, Tech Policy Press reported.
There are some signs that the data center industry is beginning to respond to the backlash. Last month, Microsoft, with a boost from the White House, rolled out a set of commitments to be a “good neighbor” in communities where it builds data centers. In response to questions on how the industry is responding to the slew of state-level legislation, Dan Diorio, the vice president of state policy at the Data Center Coalition, an industry group, tells WIRED in a statement that it “recognizes the importance of continued efforts to better educate and inform the public about the industry, through community engagement and stakeholder education, which includes factual information about the industry’s responsible usage of water and our commitment to paying for the energy we use.”
Some of the states with moratorium bills have relatively few data centers: Vermont has just two, according to Data Center Map. But Georgia and Virginia are two of the national hubs for data center development and have found themselves at the center of much of the resistance, in both public reaction to data centers and legislative pushback. More than 60 data center-related bills have already been proposed in the Virginia legislature this year, according to Data Center Dynamics, an industry news site.
Josh Thomas is a state delegate in Virginia who has been at the forefront of leading the legislative charge to put limits on the expansion of data centers. During his first legislative session, in 2024, the caucus of self-identified data center “reformers” in both the House and Senate was just three politicians. That number grew to eight in 2025, “and now, it’s 12 or 13,” he says, with many more politicians willing to vote on reform bills. His fellow lawmakers, he says, now “understand that we need to negotiate where these things go.”
Last year, a proposal introduced by Thomas which would have required data centers to perform more in-depth environmental, noise, and community impact site assessments passed the legislature, but was vetoed by then-governor Glenn Youngkin. Newly-elected Governor Abigail Spanberger, a Democrat who talked about making data centers “pay their own way” on the campaign trail, seems much more likely to reconsider this year’s version of the bill, which has already passed the House.
“I’m much more optimistic that [Spanberger] will sign,” Thomas says.
Thomas, who was not involved in shaping the moratorium in the Virginia house, thinks that a moratorium on data centers is much more likely to pass in states where the industry has less of a foothold than Virginia. Still, he says, “it’s not a bad idea.”
-
Business1 week agoPSX witnesses 6,000-point on Middle East tensions | The Express Tribune
-
Tech5 days agoHow to Watch the 2026 Winter Olympics
-
Business5 days agoPost-Budget Session: Bulls Push Sensex Up By Over 900 Points, Nifty Reclaims 25,000
-
Fashion5 days agoCanada could lift GDP 7% by easing internal trade barriers
-
Tech1 week agoRight-Wing Gun Enthusiasts and Extremists Are Working Overtime to Justify Alex Pretti’s Killing
-
Fashion1 week agoItaly’s Brunello Cucinelli debuts Callimacus AI e-commerce experience
-
Tech6 days agoI Tested 10 Popular Date-Night Boxes With My Hinge Dates
-
Business5 days agoInvestors suffer a big blow, Bitcoin price suddenly drops – SUCH TV
