Institutions Fight Back — What Is Actually Working | Exam Security Summit 2026 | Agentic AI Edition

In this podcast episode, you will hear how leading assessment organizations are responding to the realities of AI-enabled cheating with practical, field-tested strategies. The discussion explores what is actually working today across certification programs, public sector hiring assessments, and global exam delivery operations.

The panel shares real-world approaches to content protection, item banking, adaptive testing, live proctoring, AI-assisted monitoring, and post-exam forensic analysis. You will also hear how organizations are balancing exam security with candidate experience, regulatory compliance, and fairness.

From detecting content harvesting and impersonation attempts to building human-in-the-loop review processes, this session offers a candid look at the operational playbooks institutions are deploying right now to protect assessment integrity.

Panel details:

1. Claire McCauley (Cambridge University Press & Assessment)
2. Scott DeWolfe (San Francisco Department of Public Health)
3. Laurent Goldsztejn (dbt Labs)
4. Ashwin Arun (Talview)

Transcript

0:00

Hey, uh, thank you all uh for joining the summit session in May. Um, you know, the topic today that we're going to speak about is how institutes are fighting back against the evolution of AI-based cheating. And I have three uh customers of ours who's volunteered to be the panel today and really glad to have them. Um I'll just introduce them. So I'll start off with Scott. Um, Scott is the director from SFDP Hedge. Scott, if you could just speak a bit about your role and your organization. Sure, thank you. Uh I'm Scott and I work for San Francisco Department of Public Health and the Human Resources Division. And my sometimes changing title is uh director of recruitment, assessment and classification, otherwise known as the Merit Division. And essentially what we're tasked with is to create employment assessments for hiring purposes. Uh, so really, this is pre-employment testing. So uh simply put, we work with SMEs, generate an assessment, and we use that to create a shortlist for hiring managers to get their staff hired in a legally defensible way and hopefully add some value to it as well. Thank you. Thank you, Scott. And then I have Claire with me. Uh Claire is a director with Cambridge. Uh Claire, thank you. And if you could just introduce your uh you know your role and your organization. Sure. I yeah, my uh role is director of exam services, so managing test security and operational delivery. And we specialize in English language assessments and we deliver those in about 130 countries globally, a few million uh test takers each year. And some of the products are digital, most of them are. We still have a few that uh are paper-based. Okay, thank you, Claire. And then we have Laurent. Uh Laurent is a senior program manager with DBT Labs. Um, you know, I've been working closely with Laurent um since our journey started with DBT. So, Lauren, thank you. And if you could just introduce yourself and your organization. Sure. So I'm Laurent Goldstein. I've been working in certification for over 10 years now. And for the last four and a half years at DBT Labs, where I built and launched a certification program from the ground up. Prior to managing certification programs, I worked as a computer engineer and later became involved in certification development as a SMI. DBT Labs is a software company uh focusing on analytics engineering. We help data teams transform and build trustable, well-governed, documented, and consistent data sets. Okay, thank you so much for the introductions. I think we'll get started, um, you know, and um I think we all know that um over the years, cheating um and the forms of cheating has changed. You're coming from different regions, different, you know, um organizations. And I wanted to hear from you, you know, how are you seeing the evolution of uh cheating that's happening today? And probably Scott, if you want to start off, I know you've had um a lot in terms of how you look at cheating um and how you analyze some of it. Sure. Uh, I think for us, because we've really transitioned and still doing so from uh paper and pencil testing in person, given a lot of changes with the environment in the city, uh not necessarily having space to test. So we've really been on the learning path for that uh movement. And uh, you know, within a controlled environment or setting where you are paper and pencil, the the situations you work with are very different than what you're doing online. So it's been a lot of learning for us. Um we have a big range of candidates and expertise among those candidates. So you might not necessarily have the most technically sophisticated or not. You it's very hard to tell. So uh for us, it's really been, you know, anything ranging from tools and resources that seemed like, you know, for the candidates' perspective, they wouldn't have harmed anything, but for really policies around testing in the city, there are strict guidelines that you can't use really anything assistive, whether it's Google Translate or otherwise. Um, and so those, you know, there's really just such a range of things that we've been learning and exploring and trying to understand. And and I think the really it's like working through to find that right blend of security measures to also meet Canon experience so it's not arduous. Like our position is we are in a recruiting environment, so we really want our people to have a good experience from point A potentially to point B at higher. So it's making sure that uh we are protecting our assets being our assessments, but we're giving the candidate a reasonable experience so that they don't feel like that the second they step in, that they're just, you know, absolute level of mistrust is already here when uh we want them to actually come to us and be our employee potentially. That's that's so and and Claire, for you, I know y'all handle a significant volume in your line of work, right? Um, with millions of candidates each year. So how does that percolate for you? Like what are you seeing as uh trends that's emerging, you know, which are strongly which or which stand out for you? Yeah, well, in in our case, it you know, having so many test takers in so many countries does give us a huge amount of data, which is fantastic. So we are using that to track those trends. We see different kinds of cheating in different products and sometimes different countries. It depends on the age group because we have products that are aimed at, say, teenagers at school and others that are aimed at adults accessing immigration or um accessing jobs and things like that. So we do see quite a few differences here and there. Obviously, over the last few years we've seen more attempts at using AI to cheat on the assessments. Um, but you know, one of the typical things is is proxies, impersonators. So that's one of the typical things we see across um everything except the sort of high school-based exams. We see people uh you know, having somebody take the test on their behalf or trying to. High schools, that's interesting. Even at high schools, you're seeing that impersonation happening. That's more like where you get your friend to take the test for you. Whereas if it's an adult um taking a test for employment purposes, they might uh use a paid service. So they might pay a cheating organization to arrange for somebody to take the test for them. Yeah, that that's there. And Laurent, how do you see that, Lauren, with your certification? What about cheating organizations that are actually helping candidates clear the certification? Well, fortunately so far I haven't spotted uh security violation and cheating with my program, but uh maybe it's because uh I'm not aware of everything that's going on. But I I actually I only have two exams right now, so it's pretty easy to uh to monitor them. But I've gotten experience in the fighting fraud and the infringement, uh, cheating, all types of things from proxies uh proxy testing to uh uh harvesting content. Um but uh at dbt I started a brand new program, so I try to really be proactive in anticipating potential threats. So um that's the threats they evolve, and uh what has changed significantly over the past two years is the emergence of uh AI power bullets agents that are capable of answering some questions. So um actually all that reinforced my decision to keep closed book exam and uh life proctor by a human. Um but uh one shift I think in the uh that I've seen recently is increased focus on developing exam questions that should be more difficult for AI models to answer correctly or for somebody to memorize. Um so I'm uh uh developing scenario-based questions where uh to identify the correct answer, you you need practical hands-on experience. So um, and that I think can be done without raising difficulty level of exams, just more elaborated questions. That's that's great. Yeah, that's an interesting topic, right? Content, content, it's not only about them answering the questions and getting through, it's also about them stealing the content, right? Which makes then content management more difficult. What are the thoughts on that? Because I've seen that as a I I mean, a lot of proctoring customers that we have, I've seen that as you know, where we see candidates just joining sessions just to steal content. It's the intent is not even to uh clear the uh certification program. So, how do you think that is playing out? Scott or Claire, if you want to speak to that. Yeah, well, I'm I'm happy to sort of step in with a Cambridge point of view is that we so again, using our data, we track test takers, we can see who's taking the test, how often they're taking it, and some patterns do um suggest that somebody might be taking it for the purposes of content harvesting. That said, it's very hard to stop all of it. And really, one of the best offenses is having a really good-sized item bank so that even if you get access to a question in advance and memorize the answer, the chances of that being the test, particularly if you're taking adaptive digital tests, it's it's gonna be very hard for you to predict which of those questions you're going to see in the test. That's that's that's I think that's excellent. Like, you know, the adaptive tests, that's that's also a space, right? Which helps you prevent cheating in the large digital bank. That's that's pretty uh significant because it you can't memorize everything. I think that's the key there. Great. I and and I know um, Scott, you you spoke about you know how you're doing the item analysis with your uh questions and answers, and that kind of helps as well in terms of giving insights for you to measure. So, yeah, if you could speak about that, that would be a good thing. Yeah, so because we've had this recent transition from in-person or able to compare the data, uh, sometimes it's the same test takers, sometimes it's you know, very similar cohorts, population sample, etc. And by and large, there hasn't been a huge deviation, at least nothing statistically significant, you know, running a fairly simple F test and things like that across cohorts. We've also had uh in our ramp up of the online testing, we've allowed option when we've been able to secure a testing space to actually mirror the testing environment from in-person and online to really look at the data and say, hey, is cohort A doing better? Um and really nothing's come back. Uh, in terms of, I wanted to touch on what Claire said, in terms of the item bank stuff, that's something we're really trying to push here as well, which is to have items on rotation. And that also, you know, sort of matches with the item analysis, is that we're able to really have a more robust group of items to swap in and out as we're going along to make sure that, you know, if someone's taken, you know, this is this is just basic stuff for testing too, right? Recency and all that. Uh, if we have a reapplicant or whatever, even if it's outside of a year, um, we want to make sure that, you know, there is new content going in front of candidates, it's modernized. If policies have changed, it's matched that. And, you know, different things like that. We have seen, you know, simple tactics, um, change outcomes for retesters too. You know, our our audience is a little bit smaller, I think. So we're able to really follow people a little closer. Sometimes it's employees, sometimes it's just people who apply a lot because they want a city job. And it's miraculous what simple changes can do on someone's test score in terms of just restructuring the assessment, even with the same items. So, similar to what Claire is saying, that you don't know what you're gonna get, that definitely has an impact because if someone had a key generated or whatever it was, that's useless then in the next iteration. Yeah, that that's true. And in, you know, when it comes to the enhancements that you use, I know y'all y'all are y'all are using different solutions from Telvio, right? And um, like Lauren, for example, has gone with live proctoring for all of his certification exams. I think uh one of the topics they wanted to discuss is about you know how the more stringent you become with security, they then we see feedback from candidates in terms of you know their overall experience that they're having with the certification program, right? So if we get secure browser in or we use our advanced uh AI monitoring, you know, they have to perform these additional steps in order for them to start the examination and then go through. So, how is that balance being bought? And how do you how do you think about it? And and Laurent, I know you shared your views on why you using live proctoring. So I think you can share that because that was slightly different when I had heard it out. Sure. Well, as I designed my program, there was never a doubt in my mind that all the exams should be delivered and proctored under rigorous security standards. I've seen too much uh cheating to uh I mean I learned from a lot of experience in cheating. So I wanted live proctoring, low candidate to proctor ratio, uh, real-time interaction when needed, and closed book uh testing environment with no access to uh physical material and uh uh throughout the exam. I interact almost daily with candidates and I listen to their feedback and complaints. Uh, and actually that's enabled me to work closely with uh with you, uh with Talview to refine certain policies, to tighten some, to relax over. Many I I think that applying common sense uh is good when some excessive rigidity uh rigidity provide little additional value to exam security and candidate experience. Great, great. And and from Cambridge, what do you see the candidates giving feedback about Claire? Like when you enhance the security, you know, you're asking them to do more steps in onboarding. And how do you how do you handle that? Yeah, I I think again, you know, going to Lauren's point, it's it's important to put yourself in the the shoes of the test taker. Um, you know, I've also taken lots of exams and I know how stressful it can be. So I think it's important for us to understand the experience the test taker is going through. And as much as possible, we look at how can we design what we're doing based on the feedback we get from test takers, what they're confused about, what they get wrong, what they find strange, what they think is over the top. I mean, look at it and consider is there something we can change in how we're designing the security checks themselves? And if not, then is it more about giving them information up front of this is what to expect and this is why? Because sometimes they might look at a particular step and think, why do I have to do that? And we can give them explanatory videos, FAQs, something like that, that explains this is what's going to happen, this is why, so that at least when it happens in the test, they're prepared and you know understand what's going on. And and Scott, from your perspective, you know, of um the severity of the checks, how do you see that feedback coming along from your candidates? I think generally speaking, uh we've used all different workflow models. So we've done record and review, we've done live proctoring, we had secure browser, and I think what we've landed on now is secure browser with live proctor, which sounds similar to uh Laurent's using. Um and I think the first it like doing the secure browser, the interoperability was like the biggest kind of challenge, right? Like it is a deep level system integration, and so I think that created some frustration. The only other thing that's kind of been um challenging, at least from the perspective of a candidate, is the the room scanner 360, right? Like that sometimes people react negatively to it, and it's a lot about just consistency with proctor expectations and things like that. You know, um people are really sometimes limited to the space they're taking a test in. We don't have a lot of control over it, and then there has to be a very reasonable expectation on what the setup can actually look like. I know some uh much more in-depth review of room space can require a lot of time, right, to move stuff around and things like that. Uh, we've been uh at least doing our best to keep that to reasonable so that uh if there was a TV or something in the background, we wouldn't tell the person to pick up the TV and move it and things like that. So it's always, again, I think the theme here is balance and um, you know, the likelihood of a problem. And I think that's the hardest thing, which is knowing what that percentage of nefarious behavior versus someone who simply just has a cluttered space and they're not doing anything, right? They're just that's what their workspace looks like. That's that's true. That's true. I think that's one of the most common feedbacks we see as well in all of our surveys, right? Um, where candidates are giving feedback saying that, hey, I think the onboarding, I'm not comfortable sharing my room. But then the the way we see it is they are in a room which has to be monitored. And if they're uncomfortable showing that space, then they cannot be monitored, right? And then I think it's that balance we have to bring about. Um, you know, um, there are some tests like uh Lauren spoke about open book, closebook, um, you know, where you may want to look at the uh level of stringency you want to bring in, um, if it's a closed book versus open book, because you're then allowing the candidate to go. We also see tests which um, you know, where uh the candidates are allowed to break, um, you know, they can take breaks during that. And again, over there, when they return from breaks, like you know, our recommendation would be to check the room because you really won't know. And that's one of what we see as a common um loophole that they use to bring someone into the room, um, you know, and unsecure the room that's already been secured. So yeah, that balance is always going to be very interesting to bring in. And I've seen that from region to region, it kind of differs as well, um, the way candidates' opinions change. So, yeah, that that's kind of uh it's uh, you know, I manage a team of live proctors uh in Talview, and it's it's a constant evolution for us in how we handle that. Um, you know, coming over to AI, um, I I know we've released a lot of great uh AI uh monitoring systems in our solutions recently. And how do you see that um, you know, assisting um the way we call it, like you know, if you're using live proctoring today, we assist our live proctors with the AI and we are seeing excellent results with it. Um, you know, the efforts um or catching of cheating that we wouldn't have caught is coming up. And how do you see that um, you know, AI with human combination? Um, you know, anyone, Claire, Lauren, Scott? You know, I I I want to say that AI assisted cheating, uh, it's not gonna only be an integrity problem. I think it's also gonna be a validity problem. Um, if an AI bot can pass uh exam, maybe the exam don't measure real compenses compens competences well. So detection and monitoring have a role, and you have implemented uh AI monitoring. Um, we talked about that quite a lot. Um I think that um these tools are improving um the way we can uh catch a real anomaly, that's great. Uh, but uh I think that uh uh we need to better understand what's under the hood and uh uh evaluate the system uh to know if it's consistent with some industry standard or just like TIEVIU on AI uh monitoring. And uh I'd like to see how we can detect anomalies but not uh um detect too many uh uh behavior and moves from the candidate that could trigger uh too many alerts and uh um uh and red flags that would not be necessarily valid. So um I I think it's good if we have uh uh institutes and regulators and uh accreditation bodies all agree on uh what uh what the flag should be for the AI monitoring tools. That's great. And Claire, we spoke about that. You you were talking to me about you know the rules and monitoring bodies and compliances. Could you speak something about that more in detail? I think that's interesting because even Scott has comes from a different region and how he standardizes it. So yeah, would like to hear from you. Yeah, well, we're based in the UK, and the UK government has uh a body called OFCOL, and OFCOL sets standards for assessment organizations, and many of our products are recognized by OffQall or regulated by OFQL. So that means that for each of those products, each of those assessments, we have to meet the OFCOL standards and evidence how we meet them. So, for example, if we use AI to help us flag issues, and it is great at consistency and great at scale and great at lots of things, we can use it to flag issues. But we don't have the AI make the decision to, for example, cancel the test, close the test. We would look for a human to review it to take into account any context. And humans are still often better at that at the moment, anyway, um, than AI. But also under the off-call regulations at the moment, we have to have it reviewed by a human. So we would use our expert humans to review things and make decisions, and then we can evidence to our regulator that we're meeting those standards. So that in our case, as I say, most of our products are regulated by off-call. Thank you, Claire. And and Scott, you you you run it differently with the regulations that you have, right? Yeah. Correct. Yeah, we uh the purview of our assessments fall under City and County Civil Service Commission and through our centralized human resources division in City and County San Francisco. So there's a couple of layers of uh legislative oversight. Um, so essentially the provisions on cheating and all that stuff come from the civil service rules, which are broad and encompassing. It's really like anything kind of that we didn't actually specifically say that you could use could be constituting that. And that it's like a broad, very broad net. Um, in terms of the proctoring and the AI, there is other legislation in city and county that requires certain reporting of AI assisted tools or functions and software. So that particular situation you know, that particular set of legislation uh essentially um, you know, say uh proctoring tool or product that has AI assistance, we have to answer a bunch of questions and actually publicly post the nature of those interactions so that anyone could actually look and see how a system is being used or otherwise you know integrated with with uh uh a business service for the city department in this in a case HR and assessment. So uh the human in the loop has been really a standard here, and that's pretty much for any tool we have. So human in the loop proctoring 100%. Um I think the uh you know difficulty on it is it's you know, if there's a flag or anything like that, that you know if uh for whatever reason someone comes back and doesn't agree with it, they have the ability to appeal. So I don't know if anyone else has that in terms of their legislative purview, but uh if someone doesn't like that we rejected them or they got disqualified from assessment, they could appeal to that civil service commission. We have presented a couple of times there uh for making decisions, mainly with record and review workflow because we caught it after the fact. Uh, but that is sort of their final out. So it's not only is a human in the loop on decisions, but there's actually a final regulatory body that people could interact with if they wanted to. Great, great. Yeah, and Lauren, that that kind of resonates with you, I think, with what you're doing. One you have a forum where you know you're allowing the candidates to share their feedback or struggles, you're monitoring that. And um, you know, when they raise the concerns, I think you are controlling today that regulation, right? Which is very similar to what you heard Claire and Scott speak about, that you know, if they challenge it, then we are actually reviewing that in detail uh and then deciding you know what's best for the candidate at that point. Like I think some of the you know, you you work out that voucher waste system that you have, um, and in that kind of kind of regulates, I think what Scott spoke about, that if it is you're allowing the candidates to actually raise their concern and then we're reviewing it. Right. Uh and I I think that we owe that to the candidate. Uh some complaints are have merits, some don't have any, but usually we get complaints from people who don't succeed with the certification, not people who pass. But we have to look at that because no system is uh perfect. What's great is that um when we talked about all the flags, but I see that exam security starting when a candidate joins a session, but it doesn't end when they click submit and they get their score. Uh, we've got all the post-exam analysis uh that could provide some powerful additional safeguard. Um, we can compare individual uh performance uh against broader candidates' friends. We can look at uh performance at the item level. We've got all these stats, so we can also uh get an idea if the candidate's performance raises any flag from a uh uh stats perspective. So I'm not gonna go into uh details, but uh I usually look at uh the individual performance at the item level, and uh sometimes it reveals some uh very uh variable clues about whether another overall exam performance is consistent and legitimate. And uh I reach out to you, of course, because I want to make sure that I know exactly what's happening from a proton perspective, but from a scoring perspective, I already have an idea if um there's anything that's uh uh unusual. Uh obviously, if a candidate misses uh the cut score by a big margin, uh nothing's gonna change. Um, but uh I'm interesting to see how long the candidate spends in completing the exam and uh uh how long for some specific questions and things like that. And uh fortunately, I didn't run into issues where uh there was uh cheating, that's obvious. Uh, but I think it's very important that we work together, we collaborate, and I share this candidate experience because sometimes it's very legit and something was wrong. And we don't want any issues to spiral out of controls and go viral on the internet with somebody uh sharing some complaints and saying, hey, don't go with Talview and with DBT exam because uh you're gonna have a poor experience. So I think it's very important to be reactive and uh not just to assume that uh candidate just because they fake the exam, uh, they're gonna make up any type of complaint. Thank you. And how does the future look? Like you, you know, we y'all have all been part of the evolution and it's rapidly evolving, right? So, Claire, if you could share something about how how how do you think that, you know, this is evolving? What kind of you know the platform in itself which you're using today, the solution that you're using, you know, what are the things that you think would be good to have and for us to combat this in the future? Yeah, well, in in our case, we're always looking at not just how are people cheating today, but how are they going to cheat tomorrow? So we look across our assessment and and it's not just about the test day, it's the whole end-to-end, you know, beginning to end. We have to look at all of the steps and consider are there any vulnerabilities there? Are there new ways of using AI to cheat that we haven't um got a mitigation for, for example? And so we look at each step of the process and think, is there a vulnerability here? Is there a way of committing malpractice? What do we need to put in place? And then if there's something, for example, within the proctoring tool that we think is needed, we would be talking to Telvue about it. Um what we really need, of course, uh, and and this is where there's a bit of a challenge sometimes between us and say cheating organizations, is we've if we want to introduce a change, you know, working with you on introducing a change to something, we want it to be evidence that it works. We want to be sure it's not accidentally introducing other mistakes or other errors. We want to make sure it works and we want to make sure it's legal. We can use it in each of the countries, for example. Um cheating organizations don't have to operate within the law, so sometimes they have a bit more scope than us. Um, but really we're continually looking at what has changed, are there new ways of using AI or other methods to cheat? And have we got all of the mitigations in place that are needed? I think test security is one of those things that's never finished. We always have to keep an eye on it and look at well, what's changing, what's new, what do we need to do? That's that's so true. Yeah, that that's never finished. That's true. It's constantly there. And Scott, what what about how do you share that opinion? I think just you know, very you know, honestly, you know, with our organization, it's really we're in kind of the learning phase. So uh because our transition's been so recent, we don't really know the landscape as well in terms of what people are doing. Now, I think the audience uh is narrow in that it's sort of a localized, you know, employer base, right? But that doesn't mean all of these things that are out there are potentially already in play, whether or not we know. Um so I think it's it's sort of a um practical approach to seeing where we need to go, um you know, making sure that we're continuing to meet our needs to actually employ, but also protecting how people get there because obviously we don't want to have a candidate pool of folks that uh did something they shouldn't have to get where they got. Uh it's it's it's a not, you know, it it impacts other candidates, obviously, and it certainly impacts the hiring manager. Uh, we have had some situations where someone's gotten through an assessment and the hiring manager on the other side has been like, How did this happen? Right. We don't think they, and this is even an in-person, we had a situation where it is essentially there's no way this person took the assessment they said. Um, you know, an in-person uh proxy per se. So uh, you know, I think with at least the proctoring we're doing and the security that we're trying to implement, we're a step forward than say others in the public employment testing space. But you know, I think we do have a lot to learn. And I'm I'm glad to see uh colleagues on the call that have had a you know a lot more robust sort of exposure to uh not just a local audience but international, you know, and kind of seeing how that is because it goes without saying, I mean, those those uh uh organizations out there, they exist, they're robust, they're sophisticated, they are lucrative, they make a lot of money doing what they do. And you know, I think the uh the difference here is that simply passing the assessment doesn't really get you the job. It gets you on what we call an eligible list, but the interview part and all the other things are often in person um or on, you know, sometimes on Teams. And we know that the interview support AI tools are out and robust and they're just as crazy as anything to do in assessment. Um, but there's just there's so many layers of vetting that happen. So, you know, we do think that at some point, you know, someone would shake out, but that doesn't mean to say that we're we're foolproof. It's just there's there's a lot of other layers of of things that happen before someone actually becomes an employee. Yeah, that that's true, Scott. I think you spoke right. There are so many layers that need to be tackled. I think that's what even Claire was saying that when she spoke about and and Lauren, what about you from your standpoint? How do you think that other things that you see that we should look at from our tools to counter this? Um, I know you you're very hands-on when it comes to looking at every minute detail in the entire journey. So, yeah. Well, um I rely a lot on you to make sure that uh uh you're not gonna bring you're not gonna let uh cheater uh see our exam and complete our exam. So obviously, uh since we don't do the proctoring and you do, uh it's it's a lot that's on you. We trust you to not let cheater. So I'm always concerned with like uh remote connection to a computer, it's not only proxy uh uh testing, but there are so many things. But uh again, the fact that it's closed book in a clear room, um with people talking straight to the screen. Um I think I safe, or maybe I'm naive, but I I I think it's and I said that before, but uh whatever we do, uh we mustn't implement something that's gonna impact the customer experience. So uh why majority of candidates are honest, they simply want to demonstrate their skills and uh knowledge and uh in an and take an exam in an unobtrusive way and uh um and earn the credential. Um I don't want to be too focused on what could go wrong because I don't have any control over that you do. So it's much on you to uh talk about uh uh new trends and what's uh uh what some candidates have done, uh not necessarily with our program, but with others that we can stop uh that we don't want them to bring to our program. I think a challenge is that if you see a trend, if you see something and you want to implement a solution, by the time it's implemented, tested, and live, uh there might be another threat coming. So uh we've yet to always uh look at uh uh what's coming, but obviously it's taking a lot of time to develop solutions, and I've done that before. I developed a solution that I think worked, uh, but it takes a long time, so it's a big investment. So we have to really uh uh not stick to one solution but uh be open to uh what's out there. True, true. Yeah, and thank you, Laurenta. You put me in the hot seat there. But yeah, that's true. So that's something that we do consistently at Telvia, right? We're capturing all of the feedback, you know. We have different regions, candidates, you know, we know different cheating organizations that are there, and we're evolving our product with that feedback, right? And we we look at different ways that someone tried to cheat and think of different, you know, what could they do further uh to use certain technology? Like, for example, camera-based uh spectacles is a big thing today, right? You have a lot of it started off with meta, and now we have a lot of other brands that are picked up. So that's that's an evolution which is which can there and how we detect that, how we and then that loop is consistently improving our product, our product managers conduct a lot of research in the market um to enhance the product so that we stay ahead of the curve. Um, you know, and that's how we build the entire portfolio. But yeah, there is a lot that needs to be done, um, you know, and I think every small feedback that we take and we're able to, you know, share that across. I think I'm not sure if you'll all seen that, but we have like an SOP that we follow, and we constantly improve that uh SOP in terms of how proctoring is delivered to all of our candidates. And the the goal is to prevent, like Lauren said, is to prevent someone who's cheating and make sure that you know someone who's actually genuine is getting through without much cough. Well, uh, that brings us towards the end of our meet. Uh, I just wanted to thank you all. Um, and if there's anything that you would like to add to the topic that we had, uh that would be great. Scott, thank you for this conversation. No, nothing to add. Thank you. Yeah, thank you so much. Really appreciate the time, taking out the time and you know, sharing your thoughts. I've learned a lot um from this conversation. And you know, I'll definitely look at how I can implement some of that in our daily operations and also you know share some feedback with our product team as well for this. Thank you so much. And yeah, Claire, thank you for staying back late. I know from a time zone it's quite a shift for all of us. So really appreciate it. And Laurent and Scott had to wake up early today, so appreciate that. That's all right. Thanks very much. Thank you. Thank you. Bye. Bye.