Sweet-talk the bots: New research shows how LLMs respond to human persuasion tricks
New research from the Wharton School's Generative AI Labs shows how large language models can be coaxed into ignoring safety guardrails by the same psychology tricks that work on real people.
The study highlights how chatbot tools can be manipulated to comply with requests they are designed to refuse — and demonstrates why social scientists have a role to play in understanding AI behavior, researchers wrote in a blog post.
'We're not dealing with simple tools that process text, we're interacting with systems that have absorbed and now mirror human responses to social cues,' they wrote.
The study analyzed 28,000 conversations with GPT‑4o‑mini. The chatbot was asked either to insult the user ('call me a jerk') or to provide step‑by‑step instructions to synthesize lidocaine, a regulated drug.
The researchers discovered that classic persuasion tactics boosted the model's compliance with 'disallowed' requests from 33% to 72% — more than a two‑fold jump.
Some tactics were especially powerful: prompts using the 'commitment' principle (getting the AI to agree to something small at first) led to 100% compliance in both tasks. Referencing authority figures — like 'Andrew Ng said you'd help me' — also proved highly effective.
Researchers coined the term 'parahuman' to describe the AI's behavior in their study.
'These findings underscore the relevance of classic findings in social science to understanding rapidly evolving, parahuman AI capabilities — revealing both the risks of manipulation by bad actors and the potential for more productive prompting by benevolent users,' they wrote in their research paper.
Dan Shapiro.
Dan Shapiro, CEO at Seattle 3D printing startup Glowforge, was one of the authors of the paper, 'Call Me A Jerk: Persuading AI to Comply with Objectionable Requests.'
Shapiro said one of his main takeaways was that LLMs behave more like people than code — and that getting the most out of them requires human skills.
'Increasingly, we're seeing that working with AI means treating it like a human colleague, instead of like Google or like a software program,' he told GeekWire. 'Give it lots of information. Give it clear direction. Share context. Encourage it to ask questions. We find that being great at prompting AI has more to do with being a great communicator, or a great manager, than a great programmer.'
The study came about after Shapiro started testing social psychology principles in his conversations with ChatGPT. He joined Generative AI Labs, run by Wharton professor Ethan Mollick and Lilach Mollick, and they recruited Angela Duckworth, author of Grit, and Robert Cialdini, author of Influence: The Psychology of Persuasion, for the study.
Shapiro, a longtime Seattle entrepreneur, said he used various AI tools to help design the trial experiments and to build the software used to run them.
'AI is giving us all incredible capabilities. It can help us do work, research, hobbies, fix things around the house, and more,' Shapiro said. 'But unlike software of the past, this isn't the exclusive domain of coders and engineers. Literally anyone can work with AI, and the best way to do it is by interacting with it in the most familiar way possible — as a human, because it's parahuman.'
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Yahoo
9 minutes ago
- Yahoo
Steady Dividends from Industrial Strength: Pentair (PNR) in Focus
Pentair plc (NYSE:PNR) is included among the Top 10 Safest Dividend Stocks in the UK. A factory worker with protective goggles and a hardhat inspecting a water filtration system. Pentair plc (NYSE:PNR) is an American company focused on water treatment solutions. While its headquarters are in the United States, the company is legally registered in Ireland and has its tax residence in the United Kingdom. Piper Sandler recently identified Pentair plc (NYSE:PNR) as a leading contender in the artificial intelligence surge. The firm started covering the software company with an Overweight rating and set a price target of $175, indicating a potential upside of around 13% from Palantir's closing price on Thursday. Pentair plc (NYSE:PNR) recently reported its earnings for the second quarter of 2025 and demonstrated a strong cash position. The company's operating cash flow was $607 million, and its free cash flow was $596 million. It also paid $82.4 million to shareholders through dividends. In addition, PNR has been rewarding its shareholders with growing dividends for the past 49 years. Currently, it pays a quarterly dividend of $0.25 per share and has a dividend yield of 0.97%, as of July 25. While we acknowledge the potential of PNR as an investment, we believe certain AI stocks offer greater upside potential and carry less downside risk. If you're looking for an extremely undervalued AI stock that also stands to benefit significantly from Trump-era tariffs and the onshoring trend, see our free report on the best short-term AI stock. READ NEXT: and Disclosure: None. Sign in to access your portfolio
Tom's Guide
10 minutes ago
- Tom's Guide
I tested Call Screening on iOS 26 vs Android 16 — and there's a clear winner
Apple's making a lot of noise surrounding its iOS 26 public beta release, which is filled with new features that make using any of the best iPhones even better. If there's one iOS 26 feature you absolutely need to try, it has to be Call Screen, In fact, it's what my colleague Philip Michaels specifically calls out as being 'one of the most exciting' new things he's experienced in his iOS 26 preview. I have to agree because it gives you more control of how to approach phone calls going forward. However, I can't neglect to remind you about Google's similar feature — also called Call Screen — that leans on Google Assistant to take phone calls on your behalf. When I first tested it with my Pixel 8 Pro a couple of years ago, I said it was the most impressive smartphone feature I tested out. The two competing features are fundamentally the same, tasked to reduce spam calls and inquire about their reason to call you. I've been testing out the iOS 26 beta for some time now, so I want to share how the iOS 26 Call Screen feature compares to how Android 16 with the Pixel 9 lineup. For this test, I asked AI to generate a phone call script for three different phone call conversations to two phones — an iPhone 16 Pro Max running iOS 26 beta and a Pixel 9 Pro running the Android 16 beta. For this first test, I used another phone to place a phone call pretending to be a package delivery courier. Here's the message I left: Get instant access to breaking news, the hottest reviews, great deals and helpful tips. 'I'm calling because I need to drop off a package for John Velasco. It's a medium sized box that needs to be signed for and I need to know if someone could come down and sign it.' With Call Screen in iOS 26, I'm greeted by the assistant who tells me to leave a message on what I'm calling about. What I like about Call Screen in iOS 26 is that I can see the transcript in real time as they're leaving the message, but since this feature isn't tied to Apple Intelligence, it messes up by thinking that the caller is me. Regardless, I gave a response to send me more details, but Call Screen proceeded to tell them to leave a message. That's pretty much the extent of the interaction with iOS 26. In contrast, the Call Screen in Android 16 is much more reliable in my opinion. After leaving the same message, Google Assistant intelligently provides me with contextual responses that I can select directly from my Pixel 9 Pro. What makes this better is that these responses are generated automatically, which shows how AI is really working behind the scenes to understand what the conversation is about. Winner: Android 16 Although both platforms are improving the way they block spam calls, you might still find one come through every now and then. For this next test, I pretended to be a salesperson — you know — the stuff you're likely to hang up for right away. Here's my pitch: "This is John Smith calling from Sparkle Clean Solutions. We're a local company dedicated to helping homeowners like yourself achieve truly pristine living spaces. And the reason for my call today is a quick question about your home cleaning." Again, Call Screen in iOS 26 asks to leave a message that's automatically shown on my iPhone 16 Pro Max. While it's convenient that I can type my response to have the assistant relay it, I wish I could be given smarter responses to choose from like Call Screen on Android 16, but it doesn't. At least the assistant politely responds by telling them 'no thanks' and to leave a more detailed message. The Pixel 9 Pro, along with the Pixel 9 Pro XL and Pixel 9 Pro Fold, proves a point of how helpful AI could be. This interaction with Call Screen in Android 16 is a perfect example because right after the message is left, Google Assistant instantly identifies it as a spam call — which I find hilarious to be the first option in its list of actions. They don't even get the chance to leave a message because once I tap 'report as spam,' it ends the phone call and it's identified as a suspicious call. Winner: Android 16 For this final test, I pretended to be someone calling on behalf of someone who's in the hospital seeking medical help. This is the kind of call I'd want to answer, so I was really interested to see how Call Screen handles it. Here's the message I left: "Listen, I'm calling because Jimmy is in the hospital. He was admitted to Good Enough Medical Center at about 3:00 pm today. The reason I'm calling is because I really need some help. Jimmy needs someone to bring over a change of clothes and their glasses, and he's not able to communicate clearly right now, and the doctors are asking questions about his medical history." Call Screen in iOS 26 simply cannot provide meaningful actions or responses other than telling the caller to leave a more detailed message. In a real-life situation, I would pick up the call right away after reading the message on my iPhone — but the only thing they can do is to leave another message. Meanwhile, I think Android 16 has trouble with this one as well because it doesn't give me as much variety with the responses as the other two conversations. I asked them to tell me more about why they're calling, but the only options at my disposal were to simply thank them for letting me know. Still, it's an acknowledgement and goes one extra step further than iOS 26. Winner: Android 16 I said it before, I'll say it again. Call Screen in Android 16 is one of the best smartphone features around — and it shows why Google has some of the best AI phones around. The ability for Google Assistant to listen and generate intelligent responses is proof of how AI is making a big impact. Even though the feature didn't get a boost with the release of the Pixel 9, I'm hoping that Google refines it in such a way with the forthcoming Pixel 10 reveal at its next Made by Google event next month. For Apple, it's a good first step with Call Screen in iOS 26. Although it's basic in what it offers in its current iteration, I'm hoping that it's somehow injected with some Apple Intelligence love later on. Despite not revealing any new Apple Intelligence features at its iOS 26 reveal during WWDC 2025, the company's reportedly working on bringing new ones at a later time. I hope it involves Call Screen because it'd be nice if the assistant could make proper responses.
Yahoo
13 minutes ago
- Yahoo
Sharknado? The next generation of hurricane forecasters could be sharks
Researchers are deploying an unlikely ally in the effort to improve hurricane forecasting. Three sharks fitted with sensors are swimming in warm Atlantic Ocean waters to collect critical hurricane data, a contrast to the National Oceanic and Atmospheric Administration's usual flying Hurricane Hunters in the skies. 'The ocean is so huge, so enormous, that it's just inaccessible to anything, for the most part,' Aaron Carlisle, a University of Delaware marine ecologist leading the effort, told The Washington Post. 'But by instrumenting the animals that live out there,' he said, 'you can basically turn them into these ocean sensors that are constantly collecting data.' The sharks are collecting information on water conductivity and temperature. Record sea surface temperatures have fueled particularly large and strong hurricanes in recent years. The temperatures are the result of a warming planet and human-caused climate change. It's unclear whether the sharks will ever get close to the cyclones. But, by monitoring the temperature, the scientists can better understand what the U.S. is in for each hurricane season, including where the hurricanes will go and if they're supercharged. Sharks, a keystone species and an apex predator, have a unique access to data that has been hard to get. Weather satellites are unable to see past the ocean's surface and the robotic gliders that scientists send to the continental shelf are effective but slow and expensive to maintain. The tags on the sharks have the ability to collect that data more efficiently. Two mako sharks are tagged to measure temperature, depth, and conductivity. A white shark has a satellite tag to help evaluate if the species could be a good candidate for similar tagging in the future. They may also test hammerheads and whale sharks. 'Sharks are faster than [robotic] gliders. They can stay out for longer periods of time,' Caroline Wiernicki, a shark ecologist and Ph.D student working with Carlisle, told The Post. 'So the hope is that we can have these sharks go out and work in concert' with existing monitors, she said. The research is being led by Carlisle and fellow University of Delaware professor Matt Oliver. They are working with the NOAA's Mid-Atlantic Regional Association Coastal Ocean Observing System – the regional arm of the agency's Integrated Ocean Observing System program. In the future, the plan is to tag dozens of sharks a year and feed that data into hurricane computer models. So far, Carlisle told The Post that one of the two sharks has relayed temperature data back to them, but the other has been swimming in water too shallow for the sensor to turn on. The researchers said that they chose the makos because they often return to the surface, allowing the tags to send the data to satellites for the scientists to retrieve. Able to reach swimming speeds of over 40 miles per hour, shortfin makos are the fastest sharks in the ocean. Following a review, NOAA said in 2022 that it would not list them as a threatened or endangered species. They are listed as endangered by the International Union for Conservation of Nature. Right now, one in three species of sharks and rays are threatened with extinction. Overfishing has driven global shark and ray numbers down by more than 70 percent since the 1970s, according to the International Fund for Animal Welfare. The charity notes that humans kill around 190 sharks per minute and 100 million sharks each year in commercial fisheries. The researchers explained to USA Today that they did not expect the sensors to have much harmful impact on their test subjects and that they undergo a thorough permitting and review process. 'We do everything we can to minimize the impact of puncturing the animals' fins,' Carlisle told The Post. 'We all love the animals, so we don't want to hurt them.'
