Sign in or Join the community to continue
Spotting Scams with ChatGPT
Posted Sep 18, 2026 | Views 3
Share
Speakers
Jack Stubbs
Member of the Intelligence and Investigations Staff @ OpenAI
Jack Stubbs is a member of OpenAI’s Intelligence & Investigations team, where he leads efforts to detect and disrupt AI-enabled scams and fraud. Previously, he served as Chief Intelligence Officer at Graphika, where he built and led a team of open-source intelligence analysts focused on identifying and mitigating online security threats. Jack began his career as a Reuters correspondent, reporting from Russia and Ukraine before moving to cover cybersecurity.
+ Read More
Emily Hadley
Quantitative Threat Forecasting Analyst @ OpenAI
Emily Hadley is a Quantitative Threat Forecasting Analyst on OpenAI’s Intelligence & Investigations team. She works on identifying, understanding, and forecasting risks related to AI.
+ Read More
Chris Nicholson
Member of Global Affairs Staff @ OpenAI
Chris V. Nicholson serves on OpenAI’s Global Affairs team, where he uses data and storytelling to document major AI use cases and support the company’s economic research. He co-founded the deep learning company Skymind (Y Combinator W16), which created the open-source AI framework Eclipse Deeplearning4j. He previously reported for the New York Times and Bloomberg News. Born in Montana, he now lives in the San Francisco Bay Area with his family.
+ Read More
SUMMARY
Jack Stubbs and Emily Hadley from OpenAI’s Intelligence & Investigations team join Chris Nicholson to discuss how people use ChatGPT to spot scams. They explain how scammers create urgency and how sharing a suspicious message or describing a situation can help people identify warning signs and decide what to do next. They also discuss their estimate of more than 130 million scam-checking requests each week and why ChatGPT should be treated as a second opinion.
+ Read More
CONTENT & TRANSCRIPT
00:00:00 | Chris Nicholson
My name is Chris Nicholson. I'm on the Global Affairs team of OpenAI, and I'm here today with Jack Stubbs and Emily Hadley, who are members of OpenAI's Intelligence and Investigations team. Jack leads efforts to detect and disrupt AI-enabled scams and fraud. Their newly released report on scam spotting with ChatGPT estimates that people make more than 130 million scam-checking requests every week. That's more than 200 every second. So, Jack, can you walk us through … you mentioned a jury duty scam. Can you walk us through, what is that, starting with, you know, what prompted the couple you mentioned to ask ChatGPT?
00:00:38 | Jack Stubbs
Absolutely. A jury duty scam is a really insidious type of scam, actually, where criminals will pretend to be either law enforcement or sometimes a kind of court official or a sheriff, and they contact the target, normally on the phone, and tell them that they've missed a purported jury duty summons, and they have to pay a fine or sometimes a bond in order to avoid prosecution. So, it's really insidious, and it kind of plays on this very visceral feeling of fear to generate a false sense of urgency. Sometimes we see it then turn into what's known as a digital arrest or a digital kidnapping, where the victim is then kept on the phone indefinitely, while the scammers essentially control all their actions and get them to pay escalating amounts of money.
00:01:18 | Chris Nicholson
Wild. So, they're coming in, presenting themselves as the law. That's crazy. So, Emily, you estimate that more than 130 million scam checks are happening every week. What is a check? What counts as that?
00:01:33 | Emily Hadley
Yeah. A check is where a user is asking if something is a scam. Sometimes they have uploaded a screenshot from their computer or a screenshot from their phone with a picture of a text message or an email or a calendar invite. Sometimes they are typing in descriptions like, I've been having this conversation with somebody for a couple of months and they asked me to send them some money. And so, effectively, we're seeing users want to check if they are engaging in something that might cause them to lose money. And ChatGPT has been a really great resource for responding with whether the content they've shared is likely to be a scam. And it can walk them through the decision points around, you know, well, this type of scam has shown up in the news, or this language suggests that this might be a scam, and then can provide the user next steps for what they should do next.
00:02:32 | Chris Nicholson
So there's a lot of tells. If somebody can kind of pause and get away from the conversation and check on it, they can actually be advised that they're going down this fraudulent path with somebody. Jack, what do scammers do to make it hard for somebody to stop or slow down or ask for help?
00:02:50 | Jack Stubbs
Scammers are trying to create this false sense of urgency, often through fear or sometimes excitement, like a really visceral emotional response, which they can then use to essentially manipulate people into taking actions that they haven't necessarily fully thought through or actions that they might not necessarily actually want to take, such as transferring money. And what we're seeing, as Emily said, with ChatGPT is folks able to take a screenshot of a text message, maybe describe the situation they're in, share that with ChatGPT, and then get this kind of personalized, tailored, structured risk assessment where the model walks them through what's happening and, importantly, doesn't react emotionally and helps them identify what the kind of risk signals might be, what are the things here that seem suspicious or odd. And what I really actually like about it is it doesn't just return this kind of binary, yes, this is a scam, or no, this is safe. It essentially just allows people to make their own informed decision about the best way to keep themselves safe online.
00:03:45 | Chris Nicholson
So it's really as simple as taking a screenshot and uploading it to chat or copying and pasting some text that they've been sent to chat.
00:03:53 | Jack Stubbs
You really do not need to be some kind of AI tech wizard to do this. This is incredibly simple. And actually, Emily's analysis found that, I think, almost or slightly more than half of the people that are doing this are simply describing what's happening in their own words. So just typing it out to ChatGPT or talking it on to audio mode and saying, hey, this is the situation I'm in, this is what I think may or may not be happening, and having the model walk them through that structured risk assessment.
00:04:16 | Chris Nicholson
Amazing. Emily, in your analysis, what do we know about how reliably ChatGPT identifies scams?
00:04:25 | Emily Hadley
We have high confidence in this result, that it's very good at detecting scams, particularly in situations that have some of the more obvious signals. But it's also pretty good at situations that are a little bit more mixed. And to Jack's point, it really tries to give the user various points of evidence for the user to make their own decision on if this is a scam or not.
00:04:49 | Chris Nicholson
So chat really allows people to re-examine the context and make their own judgment. And it gives them agency. Is that right?
00:04:57 | Jack Stubbs
Exactly. It really empowers people to actually make their own decisions about what is the right course of action to take in order to keep themselves safe online. And, you know, when we've looked at how it performs, what we see is that it consistently and reliably identifies scams, gives appropriate safety guidance, but also we're always very clear with people ChatGPT can and does sometimes make mistakes. So it's really important to think about it, I guess, as a kind of second opinion or a second pair of eyes. And ultimately, at the end of the day, I think to myself when I'm working with AI, nothing outweighs your own common sense.
00:05:28 | Emily Hadley
I was just gonna add to that, too, that one of the things it does really well is that it can work across a wide variety of media types or input types and scam types. In this day and age, there's a lot of different scams out in the world, and it's very hard for one person, especially a person who's not often working in the scam space to be aware of all of these different types of scams. And so ChatGPT can be a really helpful interface to not have to put all that education or knowledge on the user, and instead it's ChatGPT helping them detect the signals so they can still make the decision, but not necessarily be an expert in all things scams.
00:06:06 | Chris Nicholson
Could you walk me through in detail a scam message that looks convincing or innocuous at first?
00:06:14 | Jack Stubbs
I think a lot of us receive examples of these every single day. So some classic ones are a unsolicited payment request, maybe a unexpected security alert from your bank about your bank account. A classic one I know particularly in the States are fake toll notices or fake parking fine notices. The thing is that often the actual initial message looks exactly like it could be legitimate, and if you take a beat and stop and just try to think through, do I actually know this person? Is this something I was expecting? Are they trying to get me to act with an unnecessary sense of urgency? Direct me to a link, make payment immediately. All of these are risk and warning signs that we can identify ourselves, but also what we've seen is folks using ChatGPT to take that moment to pause and think and get that second opinion on what might be suspicious or unusual here.
00:07:01 | Chris Nicholson
So Jack, are there things that ChatGPT users can do when they prompt ChatGPT to help it better assess a scam message? Are there things they should share about themselves? I know people who are going through grieving, who may have lost a loved one, are sometimes more vulnerable. You mentioned people vulnerable to recovery scams. Are there other things they can share to give ChatGPT better information to help them assess those messages?
00:07:27 | Jack Stubbs
I think as much context as you can share about the situation, the better. It gives the model more information to make an informed assessment with you as the user. But the actual use, the way to spot scams with ChatGPT is incredibly simple. It can be as simple as copying and pasting a text message, put that into chat and say, does this look like a scam? That's enough, that's all you need to do.
00:07:46 | Chris Nicholson
This was amazing. It was a great conversation. I learned a lot. I hope our viewers did too. I know that we'll come back to this theme later. Jack and Emily, thank you for joining us today.
00:07:56 | Jack Stubbs
Pleasure.
+ Read More
Like
Comments (0)
Popular
Watch More
Deploying ChatGPT at Scale: Best Practices for Adoption
Posted Oct 10, 2024 | Views 11.3K
# AI Literacy
# Technical Support & Enablement
# AI Adoption
ChatGPT Enterprise 101: A Beginner Guide to Your AI Work Assistant
Posted Sep 19, 2024 | Views 27.5K
# AI Literacy
# Higher Education
# Technical Support & Enablement