The Integrity Gap Where Exam Risk Actually Breaks Down | Exam Security Summit 2026 | Agentic AI Edition

In this keynote, you will explore why exam security failures often occur in the gaps between systems rather than within the exam itself. As Agentic AI and orchestrated cheating become more sophisticated, traditional security approaches are struggling to keep pace.

The session examines key challenges, including signal overload, limited cross-session intelligence, and fragmented risk management. It also presents a forward-looking AIOP framework for moving from detection to prediction and prevention, helping organizations strengthen trust, integrity, and credibility across the assessment lifecycle.

Panel details:

1. Sanjoe Jose (Talview)

Transcript

0:00

Hey everyone, welcome to the next edition of Exam Security Summit. Today we are going to talk about how we protect exam credibility in the era of agentic AI. I'm going to lead today's session with a keynote on the integrity gap where we discuss how exam risk is actually breaking down in today's day and age. That's going to be followed by a very exciting panel discussion by leading assessment practitioners where they are going to discuss agentic AI and orchestrated cheating, how the next threat wave is evolving. The second panel discussion is going to be headed by some of the leading experts in the exam body and certification space. They're going to discuss how they are fighting back and what's actually working and probably what's also not working. And lastly, we are going to have a panel discussion with leading identity and proctoring experts where they're going to discuss why the most dangerous cheating hides between lines. So very excited about what we have in store for you today. Hopefully, we can leave leave with a lot of leave all of you with a lot of value today and also some interesting insights. So with that, I'm jumping into my keynote, the integrity gap. High-stake exams rarely fail because of a single tool. They fail in the scenes between different tools. Most of us today think about exam security or proctoring or integrity in terms of what's happening within the exam. And then we use kind of a mismash of uh tools ahead of the exam, after the exam to cover some of the gaps. But these uh silo tools, these independent tools really lead to a lot of issues. And the today we are going to look at three specific problem statements. The first one is the data overload. So today what happens is everybody is using some form of AI in their uh proctoring, but these AI tools combined with a lot of different sources or signals which you're tracking from your primary camera, your secondary camera, your screen feed, the audio stream, often leads to a lot of signals, and it becomes really difficult for us to separate noise from threat. And that has become one of the most common challenges I discuss with customers in the last few months. So proctors are really drowning in alerts with false positives, eroding the trust. And in all of this, real fraud is able to hide behind the noise. The second issue, which is largely unaddressed today, is the fact that there is no pre- or post-session intelligence. Some of you are doing identity verification ahead of the exam. Some of you have some ways of tracking uh leak of exam questions, uh, some uh risk reports which you're probably getting from uh vendors who are probably not your proctoring vendors. But this can't really flag non-bad bad actors before they sit down or can't connect patterns across exams or candidates or cohorts. So that has been a big challenge for especially the large exam bodies. We have been in uh discussions with the recent past. Lastly, and I think the most important aspect is in spite of this entire uh spend in procuring, the entire spend in uh exam management is geared towards protecting the credibility of your exam. Most organizations do not have a unified risk management protocol. There is no dashboard, there is no control panel where you can, as a program owner, manage compliance, manage risk for your exams. Most of the decisions which are taken are taken based on anecdotal data or uh with uh more industry best practices in mind. And today we are what we're gonna talk about a new approach. Uh we call it AIOP, Assessment Intelligence Operations Platform, uh, one of the latest launches from Talview, and with that, how we are trying to also address all of these issues. But before we jump into the solution, let's also take a look at what are some of the uh costs associated with these uh gaps in the system today. So this uh recent reports say that almost 60 to 80 percent of candidates are having sessions where there are false positives which requires human review. So the fact that everything is getting flat, there is no higher intelligence which is being used to process exam uh proctoring leads to significant cost overruns because of additional reviews. Typically, because of how disparate systems are being used and risk exists in silos or risk detection exists in silos, almost four to twenty-four hours uh of latency exists between an incident happening and an actual defensible decision being taken because you need to pull together a lot of different data points before you can make decisions. And the blended uh cost of an investigation uh every time an incident happens is somewhere between 80 to 200 dollars, and that compounds significantly when you're talking about exams or credentialing at scale. And almost uh important, if not more, is the fact that one in seven candidates report feeling unfairly surveilled or accused because of all these issues. The fact that you don't have an end-to-end uh threat intelligence often leads to uh a kind of experience which is not what any of us would want our candidates to live with. And that's the challenge which we are trying to solve with an approach which we call end-to-end risk management. So the way we are envisioning this is there are four components to end-to-end risk management. The first is the before aspect, which is about risk prediction. How do you protect your exams from threat actors even before they sit down for exams? How do you identify repeat offenders? How do you identify if there are candidates from high-risk cohorts taking up your exams? And then how do you verify the identity of every individual which is come appearing in your exams? So that's that's the first pillar or the first component. The second is which all of you are very familiar with, which is how do you secure exams during the exam? And here the focus really is on how do you reduce the signal the noise and improve the signal-to-noise ratio, which is happening within an S within the session security paradigm? So, how do we uh do multi-model signal capture? How do we manage incidents instead of events and use real-time inference and scoring for this? And how do we also use predictive messaging to really bring down the uh amount of uh cheating which is happening in exams rather than just focus on detecting? And the third component is what happens after the exam, really the post-exam intelligence, where how we use patterns and cohorts across sessions and cohorts to look for useful information which can help you manage your exam risk. So it involves everything from collusion, ring detection, content exposure tracking, cross-session fraud mapping. So things which some of you I know are doing in bits and pieces, but not in a manner which is manage helping you to manage risk end-to-end. And all these four components are then laid into the four all these three components are then layered into the fourth component, which is a unified risk control panel or an end-to-end risk management dashboard, which really allows you to make decisions which are approvable, and you you have complete audit trail uh for your exam. So this is the really the architecture of AIOP, uh, the assessment intelligence operations platform. It has three uh kind of blocks within it. The first block, which all of you are probably already familiar with, uh Talvie's uh patented seven-layer security framework. So there are seven signal sources which we tap into to make to ensure that we are have a comprehensive security approach for your exams. This is everything from the identity verification of the candidate, the behavior analysis through the primary camera, the environment analysis through the second uh camera, the device security with a combination of locked-on browsers and screen mapping, the content feed analysis, the audio analysis, and the digital footprint analysis. So all of this are the seven signal sources which we tap into to build AOP. On top of it sits Talvio's proprietary living ontology, which powers Alvi, our agent D AI for proctoring. So uh here what we essentially uh is doing is rather than looking at very uh specific events like a candidate looking away or uh somebody picking up the font, which all could be false positives because it probably candidate is looking away as a part of their natural body language, or they picked up the font to put it on silent. But an actual incident is if the candidate is consistently looking at in a particular direction, which uh indicates that they probably have some source of information there, which is helping them to cheat, or the fact that the candidate typed in an answer immediately after looking at the font, which probably indicates that they had used the font to find an answer. So using combination of events to uh effectively uh predict uh an incident rather than looking at independent events and then creating uh the noise, which we spoke about, is really a core part of the living ontology. And this is built based on uh millions of sessions which Talvi has conducted, where we understand the combination of which events, patterns, or correlation between which events can be used to make meaningful prediction of an incident rather than independent events, and then use all of this uh before the exam, during the exam, and after the exam to do the pre-exam risk prediction, which is spoke about the incident management and the collusion detection of the post-exam intelligence. So that's really the first pillar of AIOP. The second pillar, and which I believe is something the uh we as a proctoring industry has really underserved the industry on is the risk control panel. So the risk control panel really is a risk management dashboard which gives you a full uh view of what's a risk you are exposed to across your exam. So you you might have different exams, you might have different cohorts of candidates, you might have different geographies. So the risk control panel gives you a view of what's your risk score uh in in a particular program or in a particular exam or in a particular risk uh cohort, which is based on all the signals which you're collecting. So who is attacking your exams from the uh cheating as a service provider group? Who is uh uh who is uh which exam is having a lot of candidates using impersonation because you're probably not doing an identity verification today, uh, where where which particular exam is where your content is being leaked in uh a variety of social forums and dark web on a continuous basis. So, based on all of these factors, what we are able to do is assign an integrity score or a risk score for each of your exams and also show you aspects like what's the dominant threat which you're facing for that particular exam and what kind of recommendations can we give you to actually address those specific threats. So this allows you to make interventions which are very data-driven rather than uh just in adding more layers of security without probably really understanding whether that you are effectively addressing the real threat which your exam is facing. It doesn't stop with that. You can now drill down with within an exam to see how what's your threat exposure across different cohorts, maybe based on geography, maybe based on other you user characteristics, which really also allows you to make interventions which are more surgical. So, for example, if if your exam is uh having a higher risk, or let's say in a particular geography, and where probably the um uh compliance needs are more liberal compared to let's say another region where you you are under there are more strict compliance uh regulations. Uh so you you your actual need is to use a second camera in the first region, in the second region, you might not even be able to use a second camera, even if you want to, but then now you have data which give allows you to make those surgical interventions, so you're protecting the integrity and the credibility of your exams. And in a similar fashion, what the risk control panel also allows you to do is to measure the impact of interventions which you're making. You added a second camera, did it actually make an impact with the uh risk score of your exam, or did it only add more friction to the candidate experience? You uh made specific changes to your content. Did it actually help you with reducing the uh uh leak of your content or improving the um uh integrity of your exams, or again, that was probably not an investment which has really given you uh fruit. So the that's the kind of intelligence which the risk control panel is uh going to uh surface to you at your end. And when we look at the uh industry, we've we are still really at the detect stage with uh many of us, but the industry is continuously thriving to move towards prediction. So we already spoke about how do we uh predict risk of for candidates even before they sit down uh with the exam. But the way we are envisioning is we shouldn't stop here. So today we are we are focused on prediction, but we want to move beyond that and move beyond that very fast because at the at the end of the day, what we want to do is prevent uh cheating. And one of the best ways to do that at a um on the ground uh level is by predictive interventions, where we have found that many candidates, especially in uh low to medium stake exams, but even in some high-stake exams, in especially in a remote environment, sometimes what leads to cheating is the lack of awareness on the rules, lack of awareness of what checks and balances are in place. So, what we are really envisioning, and which is a third pillar here, is that in in the near future we'll be able to actually prevent cheating through predictive intervention. So uh, for example, by using caution messages or nudges in flight or uh using a combination of signals to discourage candidates actively from trying to take external help. So that's really the predictive angle here. So a typical workflow, which the way we are envisioning, is uh AAOP is able to detect rising risk patterns. The candidate probably has some element of external assistance which they have already thought about. Uh, but we we would we basically use uh an estimate of their trajectory, how they are progressing in the exam, and then surfaces cautionary predictive messages in the session so that the candidate is also able to self-correct and not uh go down a path which is probably undesirable for him and for us. So, for example, a Q heads-up message like hey, we have noticed your attention shifting away from the screen. Please keep your screen on the exam window. This is a friendly reminder, not a flag. Uh, this probably uh is uh a good example of how we expect the predictive caution messages to work. So that was my keynote. The ask is uh for to the industry, to everyone here, to to um all the stakeholders here really is to unify the risk surface and move from reacting to predicting today and soon to preventing. And uh we believe that um assessment intelligence operations platform or AIOP gives you the brain which uh is required for uh for doing that, and the risk control panel gives you the dial, which uh really helps you to manage risk end-to-end, and the predictive caution messages gives you the liver. So that's really uh how we are envisioning the industry to evolve in the near future. And Talview is about to launch uh a beta of AOP in the coming months, uh, along with uh uh beta versions of the risk control panel and predictive caution messages. So we're going to partner with many of you who are already in attendance, you're who are probably already aware of some of this new uh launches from Talview, and we are going to co create and partner with uh each each and everyone as possible to build the future for the industry. Thank you.