Brainfood Live On Air - Ep405 - What the EU AI Act Means for Recruiters

0:00What is the proximity to the action when it comes to AI? So a lot of the hiring that's happening here is AI focused, AI centered, and the AI ecosystem is really sort of growing here. So the recruiters I'm speaking to here that are in that space are going gangbusters. Narrow segment though, because the wider standard work a day, we're recruiting just corporate white collar folks, that they're treading water here in the same way as everywhere else. So there's that. Okay, we're live, everybody. Welcome to Brain Food Live on Air, bringing it to you every Friday, no fail. I apologize in advance for the horrible setup I've got here.

0:51I'm basically in a hotel room and I hate hotel rooms doing sort of live streams because you always get the window in the wrong place. And it means that it's like shining in the back of side of my face. Anyway, no worries. And yes, Rob, there's actually been some weirdness when it comes to restream. I don't think it's anybody's fault. But the bottom line is, there's been a change in how... There's obviously been an update on Crowdcast in terms of how things are working. There's now live captions, which I think is excellent. But the live captions seem to have hammered, done something with Restream.

1:32So I do think that oftentimes we're not able to quite connect in the way in which we thought we would. So it wouldn't surprise me if I'm not actually going live on LinkedIn right now, which would be super annoying. But anyway, let's do some sound checks as we go. If you're watching this on Crowdcast, do let me know where the audio and visual is okay. If you are watching this on LinkedIn and you can see, please do let me know because I don't think that actually I am live there. as a result of some of the updates that have been rolled out recently. This, by the way, leads us to a very interesting conversation as to what happens when you have like a multi-vendor setup.

2:10We're going to talk about that in a couple of weeks' time when we're discussing how do you think about building a recruitment tech stack in the era where you want ideally to have the same agent having the same context. Do you go multi-agent and try and integrate them all? or do you go single ecosystem and get multiple agents within that ecosystem and hopefully remove some of the LinkedIn problems or some of the third-party problems? But yeah, it looks like we're going live on LinkedIn, which is great. Thank you, Caroline. Thank you, everybody there. That's brilliant. Okay, folks, before we kick off, we have to thank our sponsors as ever.

2:49Without our sponsors, Bring Through Live cannot continue as it is. And today's sponsor is one of our faves. It's also one of the best products out there on the market. It is, of course, Ashby. I'm here in the US. I've been here for the last two weeks. No question, Ashby is a product that everybody here respects and thinks about. It's in everyone's minds. When you talk about what ATS have you got, what you're upgrading to, these guys are always in the conversation. So kudos to those folks that have done fantastically well. They're also brilliant at community, I have to say. And I have to highlight, they're actually going to launch, I think the inaugural Ashby One is happening in London next week.

3:35so I do believe if you are in London and you're a TA person and you might want to get involved in a great invite only conference click on the link I've just shared there and just join the waitlist or apply to attend I do believe it's free but you do need to apply as a waitlist of course they're going to screen you but you get an opportunity to see the latest and greatest from the product but also interact with a lot of your peers that are in industry which is always a good idea. Ashby won September the 9th, happening next week. Okay, we have Adam Gordon here. Great to see you, Adam. Wearing denim.

4:11Is that the new fashion? Is that what's happening? I don't know. Yes. I don't know. Don't ask me questions about fashion. I don't know, mate. I mean, you're a princess, Adam, so it's quite possible this is what you know how to do. I just wear whatever I pick up when I first get up in the morning. that's it i've tried to improve my my wardrobe as well i realized that you have to refresh it you can't just say i have a great shirt um and then leave it there for three four years or whatever because by that time it's been in the wardrobe for that period of time your body has changed the shirt has changed the fashions have changed no longer a great shirt so i've only got one i've only got one comment which is in my 20s and maybe sort of 30s i wore a lot of pinstripes like braces tie pocket square all that sort of thing because i was probably trying to create additional gravitas in my appearance and you know this sort of status around being maybe older than i actually was and now i wear denim and like you know fluorescent t-shirts and stuff like that because i'm old and i want to look young yeah that's exactly what happens isn't it like we we we we try to overcompensate for what we haven't got um i totally get it no doubt i've been told off as well for wearing my my my jeans are too tight on you're too old for that it's like i think you're right this is what also why i'm wearing collars more now so t-shirts are gone um like i need to dress i want to dress my age um but i don't think i want to be one of those guys that's clinging on uh to an image that isn't the case i don't have a 25 year old body mate you know You know what I mean?

5:53So that's not happening. Growing into my strength, my mature strength. I mean, this is too much sharing here, but you say you don't have a 25 year old body. I weigh myself every day. I've been doing a lot of exercise and paying a lot of attention to my diet. And this morning, I don't know where this is in kgs. Oh no, I do, it's about 89. I was 13 stone 13. That's the first time my weight has started with a 13 since I was a teenager. yeah yeah that's good mate you're looking good and you see i see you doing like the veterans rugby stuff and whatever that's that's good quality exercise because um when you're doing competitive sports um you're you're exercising and in a level you don't do in the gym because the gym is very predictable you can kind of mentally sort of uh gee yourself up to do a bunch of sets or whatever but in in a competition against other people you have to react immediately and so you're calling on your body instantaneously which I think creates a different style of exercise so anyway I need to do competitive sports folks anybody want to do white collar boxing with me let me know paddle

7:05alright listen let's get on with it let's review the newsletter from last week did you read it mate and if so what was interesting man yeah okay so let's start off with Rob Davies he's the head of resourcing at Grand Thornton And he gave a couple of insights from candidates and it's about can they trick AI screening? So he's given some insights around things that they've identified, such as a candidate who applied hidden instructions to large language models inside their CV in order to manipulate it into giving the candidate a high score. so this is a reminder that as recruiters adopt ai candidates are learning how to gain the systems and you've got it it's like whack-a-mole right you've just got to keep understanding what's going on look out for the tricks but that is absolute genius adding hidden instructions inside a cv i remember some i remember somebody doing it on linkedin and adding in a prompt to a a prompt to chatbots i said large language models that were coming to their profile anyway yeah good stuff it's it's actually combining quite an old technique with a new technology so you know back in the days when um we used to do keyword searches um and one of the standard techniques was to basically keyword stuff your cv put it in white text and put it in font too So very, very small, stick it at the footer.

8:41And basically, recruiters keyword searching on an ATS, you just come up because basically, you'd be tons of keywords that would be hidden in your CV. Same technique, white text, small font, but this time, prompt injection. That's the term, folks. You're assuming the reader of the CV is no longer human. So that for me is a big transition. You're correctly assuming that and you're trying to basically persuade their AI to prove your resume escalating forward, which is, I think, really, really interesting. But obviously, we as a recruiter, we've got to think about how we police against that. I'd be interested in know, folks, number one, have you encountered prompt injection in CVs and job applications in your experience?

9:31And if not, sort of have you got a method to detect that? Maybe they've gone through without you knowing. And if so, what are you doing about it? Like, is there a technique to basically sort this out? Let me know in the comments, folks, because I think we need to probably, we're overdue a fraud-orientated webinar. We should do a brain-through live on that, because the techniques evolve faster than our ability to report on it. Okay, great one. Have a read of it. Give us a few more, Adam. So Bill Fisher, I don't know where you linked it to. I can't remember, but I saw it on Twitter, I think. Bill Fisher was talking about, I think his headline was, The Future of Workday is Palantir.

10:16So here's what the argument's about. And I hear this a lot on some of the tech podcasts and things that I listen to. systems of record are not going anywhere. Companies have embedded them deeply. But what gives the system of record a new lease of life is the ability to allow agents to come in, query the information, bring it out, and do other things with it. So it's like almost a new lease of life for enterprise software. we have to give agents the right knowledge base and the right source of information for them to do a really great job so if you go on to chat gpt and you say give me a report about x it'll do all right but if you go on to um a system and say give me a report about x and it's got access to all of your workday information and all of your salesforce information and all of your poetry information or whatever else, right, what it produces is instantly more specific and targeted to your use cases and just vastly more helpful.

11:29Yeah, and that's what we want from these, like, vertical sort of agents, really. We have ChatGVT on the open web. That's great. That may replace search. That may replace general queries. Fantastic. But when we're coming to do work, like specific work, We want the model to be trained on our own intelligence, our own context. And as you say, it may well be that the future of enterprise software is that it's a repository for all of this data. It's essentially the information storage that creates the opportunity for intelligence that the AI then can process and pull out. Very, very interesting. Bill Fisher, by the way, good friend of ours, good friend of everyone's, really, really nice guy.

12:14But he has thankfully started to commentate publicly on his vision as to where software development and technology is going. Well worth following Bill for that reason alone. I think he's a very good communicator of that vision. He doesn't sort of handhold you. You know, he'll take you into deep water and to the technical stuff. But I think he's human enough to be able to communicate to the laity as well as the experts. I've just shared his ex profile on the chat. Go and follow him there if you can stomach being on Twitter. But he posts more on there than anywhere else. So he's worth following for that.

12:55All right. Give us one more, mate, before we get into this. So Shona Waters, who I don't know, but Shona asked whether we're asking the wrong questions or she suggested we might be asking the wrong questions about AI-generated content or any AI-generated work. And she said, we shouldn't be judging the fact that people are using AI to get text created. What we should be judging people on is whether they actually remain accountable for the final output, as in, does it truly reflect what they want to say? and is it um you know have they even check have they first of all have they even checked what it's saying before pasting it onto linkedin or onto their dissertation or whatever it is but also does it actually really reflect what they want to say and my view on this is if you have a if somebody if somebody puts out a blog or something like that and it sounds really good and then you talk to them they actually don't know anything about the subject that's kind of just fraudulent you know that's not that's not something that's that's that's like the modern day equivalent of plagiarism um but if they've just used it to shortcut getting their message across or finessing their you know um the comprehensibility of their i don't know if that's a word anyway it's ironic considering i'm talking about being clear with your communication um everyone understands comprehensibility that's um if it's not a word uh before it is now so let's roll with that yeah um so you know i i i absolutely agree with this i just think that um like calling people out for having an m dash in their you know post is probably unimportant what's more important is what's the what's the actual resonance of the uh of the of the message that they convey Yeah, yeah.

14:51It's a really interesting blog post. I really recommend people read her post on LinkedIn, actually. I think it was better, more concise, the description of her point than her bigger sort of substack. and the basic situation is folks essentially she's a great writer but she was called out by someone who alleged that she used AI to produce the post and Shauna says she didn't do that but then she used the opportunity to engage with the person and started to explore like why are we judging this and obviously judging it is we do, like we are outraged at when someone sends you if you send someone a handwritten email they respond back to you with AI gen, are you not annoyed?

15:38Of course you are. And what Shona does is trying to explore from a psychological point of view, what is going on with this? Like what is the etiquette of AI use of generative AI when it comes to communication and publication? Why are we upset at it? How do we use AI in such a way that it doesn't upset people? So fascinating conversation fascinating thing to read for me i've avoided all of these issues simply by not using ai to do content production um at least not on written form um it's just a commitment that i've i've made um it's it's it basically means my writing is worse because i don't have the help of ai um but it's also um easier in a sense i don't care about that it's like okay i'm just going to stream a conscious consciousness this um it's going to be tons of flaws because i've not actually gone through it, but it is what I'm thinking right now at the time.

16:33And that's how it goes out on things there. Anyway, brilliant blog. Go ahead and read it. And Debbie Clement has just put down an incredible comment here on Crowdcast. If you want one Crowdcast, I do recommend you read it because that is a setup for a second review live on exactly this. How do you actually set up a system that screens for pumped injections from candidates? Really good. All right. We've But EUAI Act, Adam Gordon, you actually must know something about this because you are a product vendor. You're a CEO of a company. You're responsible for this in some way. What's your take on EUAI Act, particularly as it pertains to the UK?

17:14Because we're out of the EU, but what does it mean to us? I'm actually confused on that. So your view. yeah i mean so it's it's very very rare that you frame a question to me in that way that you expect i do know something about it normally you say you and i know nothing about this adam we're complete imbeciles and we're going to get the experts on this time you've got it wrong but the other way around i know almost nothing about it and you know the reason i know almost nothing about it is because as a technology provider, I don't include any candidate profiles or publicly identifiable information or anything like that.

17:57I have put in place a whole load of different safeguards to ensure we're not breaking rules and all that type of thing. But this all relates much more closely to how we treat individuals. we don't assess people or source people or anything like that. Our technology doesn't do any of those things. There's millions of technologies which do. So I'm definitely not an expert in it. I do follow Martin Redstone and I do aim to keep up to date, but I'm looking forward to learning. Yeah, good stuff. Well, thanks for being candid, Adam, but you've already described something that we can learn from, which is the application of the technology is significant for how the EU AI Act affects you.

18:51So the certain types of technology or the certain uses of it is going to go up or down. For those who don't know, poetry basically, I mean, you want to summarize it in a single phrase, but it's basically a recruiter enablement tool helps you get better, but it helps you kind of, you know, create more marketing collateral, for instance. It helps you kind of do better research and do all of that type of stuff. It's not there to assess candidates or score candidates and stuff like this, which is where it goes into a high risk area. OK, we're going to bring on our guests here, people who might know what they're talking about, which is great.

19:23So folks who are on the gallery, I need you to get off the mute on mic. So let's do that and then we'll bring you on. Martin Redstone on first. Let's say hello there. We have Natalie Farmer. We have Alexandra Davoli as well. let's do intros uh martin great to see you a magnificent beard by the way what's happening with this um it's uh like it yeah overdue trip to the barber uh no i can't unfortunately look as uh uh as her suit and uh uh well groomed as adam um unfortunately but uh thank you for inviting me on uh good to see you guys um yeah i'm not going to go back to the fashion conversation either much to my wife's child grain.

20:04I have about 17 of these t-shirts and I just wear the same one every day. Obviously the same style, not literally the same one. I have a washing machine for that. So hello. Just to introduce myself, for those that don't know me, 21 years in the recruitment world. Last half of that has been all in on applied AI, building conversational AI systems in recruitment on both sides of the fence for vendors and TA teams and recruitment businesses. Last three years, been purely focused on governance, risk and compliance. And since the beginning of this year, been working mainly with Warden AI as head of Responsible AI, where I work with our customers in the wider industry on understanding how to implement AI safely, responsibly and ethically and legally.

20:56Great stuff. I've just shared LinkedIn, sort of Martin's LinkedIn there. Must follow on that, folks. Go ahead and do that. We've got Natalie Farmer here as well. Natalie, can you quickly introduce yourself? Who are you, what it is you do? Hi, yeah, I'm Natalie Farmer. I am a partner in a law firm called Field Fisher. I work with the London and Silicon Valley teams. They're mostly advising tech clients. I specialize in technology and data regulations and laws, which of course includes the EU AI Act among other relevant sort of adjacent laws like GDPR, DSA, Data Act and so in the recruitment space you know the AI Act has very much brought what I do into the sort of Venn diagram of tech recruitment and law because we'll talk a little bit more about it today of course but because the way that EU AI Act regulates the recruitment space or the technology use in the recruitment space but we'll get to that good stuff in a moment but that's me.

21:53Yeah, brilliant to have you on the show, Natalie. Shared her LinkedIn in the chat as well. Again, must follow. However, it's very important we don't get experts on just sort of esoteric experts. We also need practitioners also, people who are potentially implementing these technologies and trying to juggle the compliance effort, which is why it's fantastic to have Alexandra Davoli here to join us as well. Alexandra, wonderful to see you. Can you quickly introduce yourself? Who are you? What it is you do?

22:23oh alex yeah you're on mute i think oh alex i think you're on mute or at least the mic is not on even though it looks like it's okay there oh still no i'm afraid yeah maybe you want to refresh the the yeah go ahead and refresh the the the chrome and then come back and hopefully we can we can get audio yeah no worries um okay as Alexandra's getting on with that let's just recap where we're at from a factual point of view so EU AI Act has been around the blocks for several years now what happened in September, so October the 2nd, so four weeks ago something happened, what was that thing, Natalie Farmer?

23:09Well not as much happened as we thought it was originally going to, so on the 2nd of August some of the rules under the Act came into force so you're right Hung this law has been around for what feels like forever for myself I was tracking it through the legislative process which means I've been looking at versions of this for even longer and it's been excruciating we're finally waiting for these things to come online the reason that we're eager is because then we get guidance we get decisions and we can actually interpret what is a very difficult piece of legislation I think to read but the deadline for the um coming into force of the the majority of the rules and that piece of legislation which we call the high risk risk classifications for AI were deferred.

23:52So in August this year, we did get certain rules on transparency, which are very important, coming into force. But the majority of the high risk stuff, the stuff that we really care about is not in force until December 2nd of next year. So this huge amount of grace has been given to us, which is great, because frankly, I don't think we were all ready. And there's a lot of work to do. So I see this as a huge positive, as well as a slight negative because the sooner it comes online the sooner we start figuring out exactly how this is going to play out all right brilliant natalie so already been educated on this there's a kicking of the can down the road to a degree um and we're all grateful for that because it's like we're not ready um in fact why don't we have a look at the poll on the right hand side of your screen folks if you're watching this on crowdcast you should see a poll there um go ahead and vote on that.

24:44It's quite a simple question. How well prepared is your TA department for the EUAI Act? Answer, not prepared at all. Covered the basics, but worried a bit. You are compliant. You're well covered and you're 100 % compliant. You're rock solid. Do put a vote in there and let us know how you're getting on. Okay, great stuff. So we've kicked it down the road, Martin. What does that mean though? There's a couple of characters that are involved in it, the vendor side or an operators so from the vendor point of view what does this mean for them like what do they need to get in order alex how are you doing i still think we have no audio for alex i'm afraid to say i don't know why that is um needs to be on stage no i think she should be able to say something um let's add a stage anyway alexandra can you say something no i can't hear you um i wonder what it's a volume thing or something i don't know um could be a mic but alex do you have a do you have a headset sometimes um when i use different platforms i need my headset could be yeah or change the input when sometimes it's uh yeah changing yeah there's a there's basically a widget at the bottom it's kind of there's a There's a way to change the settings, the bottom three sort of dots.

26:09Give that a shot, Alexandra. Let me know how you get on. Go back to you, Martin. Vendor side, what should they have done by now? And what does this kind of reprieve give the vendors? Well, the reprieve gives the vendors some time. And also, hopefully, fingers crossed, a lot more guidance from the EUAI office. That's one of the biggest reasons why there had to be a delay is the fact that, as Natalie said, there was such a lack of guidance that came out of the AI office in time for us to be able to make sure that organisations are compliant. There's still a few things we're waiting for, CE marks and those kind of things.

26:44But we're starting to get a small kind of dribble of guidance coming out now and consultations on guidance and what have you. So right now, it's about good paperwork. It's about good audit trails. it's about making sure that you're doing you know doing your utmost to ensure that risks are mitigated bias auditing all those kind of things so good compliance and good governance practice good best practice because you know until we actually have the guidance we don't know how to be 100 compliant to the legislation we're hoping to see a lot more of that coming out over the next three to six months what it doesn't mean and unfortunately i have seen this in certain parts of the industry.

27:25It doesn't mean a cheering kind of celebration that compliance has been delayed. What it does mean is that you've got time now to get your ducks in order. Yeah. So from the vendor side, what they should all, I'm now, to be honest with you, I am seeing a pretty consistent communication from mature vendors, let's say, where they are saying they've got these buy checks in order, they've got audits in place. It looks like it's becoming more mainstream. So I wonder whether, you know, by the time the act becomes enforced or enforceable, then vendors should hopefully be already on the mainstream. I think with a lot of those vendors, especially the global vendors, they're in a really good place because much to, again, the misunderstanding in the industry, the US is so far ahead when it comes to regulation compared to Europe.

28:20We've got multiple states that have brought out very similar legislation to parts of the EUAI Act or the whole of the EUAI Act. And so they're getting themselves compliant to those regulations now, which means that they're in fantastic order when it comes to the go live, the enforce of the EUAI Act later next year. Yeah. So remind me again when that was, Natalie, when does it actually happen again? So there are three important dates. One has passed. So we had 2nd of Feb 2025, when the rules on prohibited AI came into force. So there are already some prohibitions out there. At the same time, the rules on AI literacy, which Martin just mentioned, came into force as well.

29:03AI literacy being the people responsible for AI within your organisation ought to understand what it is that you're using. they need to be able to probe it due diligence on it depending on their role they have to be literate in the context of the ai that they're using that happened in in feb 2025 2nd of august 2026 the rules on transparency so if you are interacting with ai letting a user know that that is what they're doing those rules came on on on deck and the deferred deadline which is now the 2nd of December 2027 is for the high-risk AI solutions. That is a huge amount of the law. The vast majority of the Act focuses on those high-risk classifications and what you need to do if the technology you're using is high-risk.

29:50So the majority of that legislation really doesn't take effect now for another 18 months. Right, so high-risk I'm saying is candidate scoring, ranking them and then promoting so basically it's good to go for I'm seeing a lot of those tools that do this right and I looked at the EU AI Act in terms of you know how does any of these sourcing tools fit into that because it feels like they're ranking even though you know they're not necessarily saying the ranking but you're producing a limited set of candidates that are visible to me based on my prompt so presumably you've dismissed a bunch of people that that I don't see but it looks like you can still use that technology up until um the the second date you mentioned third date you mentioned should i say uh natalie which is the 27th of which is december uh 2027 very very interesting the um alexandra go ahead martin quickly oh no i was going to say the the the the point is it's not just about kind of ranking scoring it's any system um and it's intended use as well so um you have to kind of um think about it from that perspective it's not just how you bring it to market.

31:00It's the intended use by the deployer as well. And so by deploying, you know, the employer who puts that system into use. But also it's about any system that has a material impact on a material influence on that decision. So it could be everything back all the way to programmatic advertising. So we have to think about it in any way, shape and form where a candidate is either you know brought down or brought up and so you know if you're programmatic advertising and you're thinking about you know we're looking at you know a specific criteria where we're going out to a specific job board and it's all done programmatically is that having a material influence on the next decision on the next decision on the next decision because ultimately and this is the bit that I hear more and more often is it's fine a human makes that final hiring decision doesn't really matter it's how that that list of candidates has got that final decision making process as well yeah the assembling of the list of people there Alex I hope we've got the chance to hear your voice here can you say something oh no sorry sorry Alex we don't we don't know what's going on but stay with us keep wrestling away with it um and if you do sort of uh uh get the audio back on just chime in but we'll keep you on uh uh on on on the gallery for the time being um adam you wanted to say something here um go ahead no i didn't okay so sorry so just to just to jump in on on something else that natalie was talking about so obviously august the second this year we saw certain rules coming out around transparency And we always talk about transparency when it comes to responsible AI and transparency around kind of black box hiring and those kind of things.

32:49But this transparency is a little bit different. And actually, it's very, very relevant to the industry as well, because it's all about making sure that you're being transparent when people are engaging with AI systems. And so what we're seeing now is a lot of recruitment chatbots, a lot of voicebots and those kind of things. And we have a habit in our industry of, and I can never say this word, I've got far too many teeth of anthropomorphizing. You can tell I can't say it. These chatbots. So people kind of almost want to believe that they're human and that you're interacting with a recruiter or something like that.

33:25And that's no longer acceptable. We have to make sure that we are either having a disclaimer on the chatbot, the voice bot, the avatar, whatever, to say that you're interacting with an AI system. So that's something that's very, very relevant for our industry right now. So it's live. There are certain rules around whether you have to implement that now or you can wait until December. but if I were you and I were running a recruitment chatbot or a voicebot or anything AI system wise that the public, that the candidates are interacting with, I'd make sure that you've got some kind of disclaimer on there to say that they are interacting with an AI system.

34:02Yeah and I wonder whether that's going to end up being quite a significant sort of design choice for people because there is a pathway where AI became more and more like a simulacrum of a human being um and perhaps that this type of need to declare that it isn't a human being leads people to i can imagine vendors saying look we're going to solve this simply by having a robot there literally you know create an avatar that is an a human uh robot pull back from the anthropomorphism um i have some some teeth i've got teeth problem as well um it's difficult um actually that's fine that's fine as long as it's obvious that you're that you're interacting with an ai system then you don't have to be you don't have to make that disclaimer um but you have to make sure that it's obvious as well yeah so i'm seeing go ahead we're seeing some debate at least i've i've certainly discussed it with clients is how obvious does it have to be some of the uses i think when you know you're interacting with a chatbot and you can tell from the syntax of the responses you are dealing with a software you are not dealing with a human um it seems to me that generally speaking people are erring on the side of caution and they want to make it obvious there is a sort of comment in the chat here from Sinead who's asked you know why are employers so reluctant to tell candidates and I think to Martin's point people expect human interaction in recruitment to an extent because there is an incredibly human process that you're involved in but we are seeing those asynchronous interviews now which are permitted but there has to be transparency the regulation under high risk would depend on what in the background is happening with respect to the responses as an individual that you give if it's just a transcription that a human then looks at we might not be talking about high risk but absolutely to martin's point we'd be talking about transparency if the interview you're having otherwise has the look or feel of interacting with a human and i've seen some pretty sophisticated ones which you know you could easily be fooled into thinking you're dealing with a human so there is a question also in here about uk and what happens to uk we i can mention that now we can we can talk about it uh later but i think that this is becoming relevant globally obviously the euai act uh it regulates impact on the eu market but that does it that's extra territorial in that the developer or the deployer could be based elsewhere if the candidate is based in the UK and there is no other EU element yes you might escape EU AI Act but if you are a UK business interviewing globally for a role and you're dealing with an individual who's based in Spain for example you're pulled into the legislation that way as well so it classic Brussels effect this is going to seep out far beyond the 27 member states.

36:50Yeah really good folks there's a lot of questions coming into the chat which is fantastic but we're going to lose track of those questions as we go through. So the best way to do this, can I ask you to put them in the Q &A feature on the right hand side of the column. So if you have a question, just put it in there. If you've already asked the question in comments, just copy and paste what you put, stick it in the Q &A, and then we're going to get to those questions. We want to make sure all the questions are answered towards the end of the show. So the last 15 minutes, we'll be answering questions.

37:18Any questions you have, just put them there and we'll get to them in due course. By the way, if you're watching this on LinkedIn, do the same thing. Just put your question in the comments. We'll pull those into the Q &A as well, and we'll get those answered as well. Okay. Very, very interesting about the explicity, the obviousness of AI. I suppose the simple fix is the introduction, isn't it? Like vendors can simply say, hi, I'm Hong, I'm your AI interviewer, and that will do, right? Is that enough to basically get you through that okay you have to declare and then you can just you know you could be as anthropomorphic as you wish um sorry mine uh you got it in my head now um and uh and and you should be should be good to go is that right yeah i think that probably is right i mean we haven't seen it tested yet but that would be sufficient to me to make it clear that you are interacting with ai and it would satisfy the transparency requirement i think if there was any handover you know if you're dealing with bot one and then they hand over to bot two do you have to reintroduce the concept perhaps it will depend on the the quality of what you're dealing with i think but yes i don't think the intention of the legislation here was for there to be banners everywhere saying this is ai but it was to stop deceptive practices so i think you have to kind of feel your way through where the line is there um the better the ai the more likely you need the disclaimer because if it's really really fabulous then it might trick people into thinking they're dealing with something that's real so to be cautious something somewhere a little statement at the beginning a caption in the corner there might be the development of certain symbols that we start to see emerging where you know it's ai because there's something just tweaking in the corner uh we don't know that quite yet but yes to your point hung i think that would be sufficient in the most part yeah and and and to natalie's point you know when you when you go back in in time to kind of the original intent behind the EUAI Act.

39:21It wasn't to kind of overburden with banners and this and that and everything. It was to create a sense of trustworthiness when it comes to the use of AI. So you have to kind of keep that in mind when you're thinking about how to operationalise a lot of this legislation. It's about trustworthiness. It's about protecting human rights in the way that AI processes your information, your data, makes decisions, etc. But it comes down to a core of trustworthiness. Yeah, really good. So a couple of key points. Number one, use of AI, got to make sure that the candidate knows that there's an AI interaction on this.

40:00So you can do that with the AI could tell you, or before you enter some experience, you could announce it in some way. There has to be some sort of audit trail. So you've got to basically understand or document essentially how the AI, this is from a vendor side, the vendor's got to be able to produce documentation to say this is how the system works and make that readily available. You've got to avoid, at least sort of with that deadline coming up in 2027, got to be careful about this issue of ranking candidates, even though probably nothing is going to be enforced until about a Europe plus away.

40:45So just to jump in on that, what this Act is not doing is banning things like scoring, ranking, matching, et cetera, et cetera. It's creating a level of governance around that to make sure, again, that those systems that are in place are responsible, trustworthy, ethical, etc. And so, you know, you have to have management systems, you've got conformity assessment, you've got model cards, you've got, there's going to be a database of registered systems within the EU, all those kind of things. So it's not about banning these. And I think that's very clear. You know, for me, somebody that's been in the industry for a long time, worked on AI for a long time, you know, there is so much good and positive that AI can bring to our processes, but it has to be done properly it has to be done responsibly it has to be done legally so this isn't about banning them it's about making them trustworthy yeah anybody we talk about this we're bored of this analogy but i always say that ai deployment is just like a car you know cars are inherently dangerous but we all drive them because they're better than walking everywhere in the rain if you live in london um but we have to get an mot we wear our seat belts there are brakes the legislation is exactly that it's product safety legislation tech for a piece of tech so to martin's point nothing's banned but if you think about the life cycle start to finish and we can go right from targeting a job advert all the way to recruitment final decision making is it safe and in this sense safe often means non-discriminatory non-biased wouldn't otherwise is breach fundamental in human rights.

42:21And so with that overarching sort of analogy, it's clear that you can do all of this and people should be doing it. It's fantastic. And there's a huge amount of efficiency to gain. But fairness is being baked in by the legislation here. Sorry, Adam. OK. I've got a question here on vibe coding. So all of this applies to vendors, right? But the issue of vending and software production, that's like completely diffused now. So I could just be a dude and decide to vibe code my own thing. And, you know, it's in breach of everything. Like, OK, what happens here? So I'm using it. Is there a personal use component to legislation?

43:06Do I have to sort of somehow be compliant? Is it a size of company thing? How does it all work? Does anyone know? there is a personal private use exemption if you will you know we are regulating two characters here the providers of an ai system and the deployers of that system so in the plastic sense the provider will be um the tech provider the platform and the deployer will be the business who's just using it but there is a there's an interchange here where you could become the provider of a system because you take technology, take a license, you integrate it into your own platform or service or model and then you have your own customer base.

43:51So you don't necessarily have to be the person that invented the model, who developed the code, to end up becoming a provider of a system under the act. So you have to be a little bit careful. Vibes coding is an interesting one because it essentially allows any of us to become coders with very little experience using natural language processing which is cool but it also means we can do things or we can change things we can add on to things that may be unintended so if you have some person in your business who's willing to vibe code and make something a little bit faster or whizzier and they are deploying that within the context of your business in say the recruitment context text you want to be monitoring that and taking some care of that because inadvertently they might take some relatively administrative cv passing tool that's looking for you know spelling errors um and tweak it so that it ranks the cvs or promotes certain keywords or influences then decision making or effort in the recruitment process which may start to tray into the territory that martin was mentioning earlier which is having some significant or material influence over decision-making related to employment.

45:04It reminds me of the same type of thing as GDPR, right? The companies can, like vendors can provide a certain thing that's embedded with options for different organizations and it's up to the buyer and the talent acquisition team to decide how they're going to enable those solutions and features. Yeah, so there is a liability question here as well, Adam, to your point, which is, again, we talked about the intention behind some of these systems. So they're put on the market for a purpose. If the deployer then takes them and uses them in a different way, there is an opportunity under the legislation for liability for certain core provisions to shift because you've used it in a way that was never intended by the original provider.

45:55provider. So some of this will come down to contractual liability management, which is very GDPR-like, in that you have controllers and processors who manage that relationship quite carefully. Processors do a limited amount of things, controllers kind of call the shots and run the show. It's not so dissimilar. In the AI sense, I actually think that the contract management, liability management piece is really essential because you're going to have two parties, provider of an AI system, deployer of an AI system, who are very much dependent on each other for their legal survival, if you like, because on the one hand, you are going to, as the deployer, as the buyer of the tech, the user of the tech, you're going to be reliant on that provider to give you a ton of information for your own compliance purposes.

46:44They are going to be looking at you to do only what they say you can do. There is a reputational risk as well as a direct statutory risk. So lots of information sharing, lots of work to be done to allow your customers to be trusting in what you're providing them. And you know, this grace period we now have until December 2027 for the tech in this space is really important because if you're a vendor who hasn't quite got that right yet, there's going to come a point in the sales funnel where you can't sell because your customers are going to be asking for things you don't have and that's going to happen long before 2nd December 2027.

47:21Yeah and we're actually seeing that already happening and you know and you know a lot of people are coming to us for support on that as well because it's you know they're being asked questions like you know do you do you know bias auditing and they don't and they're coming to us because it has to unlock the procurement process so we're already seeing that so if you haven't got your your house in order when it comes to compliance under the UAI Act even though you've got time you're going to start as Natalie said you know coming a cropper when it comes to the procurement process because TA teams as you saw from you know Debbie's long list of questions earlier they're getting savvy and they're asking the right questions of vendors yeah very good stuff folks I just noticed the time we are flying towards the end of this show incredible as it is.

48:06Obviously, we need to continue having conversations with EU AI to bring this sort of, I think we probably need to do a quarterly on this just to keep pace with it. But in the interim, let's not make sure that the conversation ends today purely based on us needing to come off air. So make sure you're connecting with everyone who's sharing this conversation, everyone who's interested in this conversation. Best way to do that is fairly straightforward. Grab your LinkedIn URL, just stick it in the comment thread. If you're watching this on LinkedIn directly, do the same thing, LinkedIn URL in comment, and just connect with everyone else you see doing the same.

48:45Everyone is signing up watching this because they are interested in what the implications of the EUAI Act are. And this is a good cohort of people to share knowledge with. I think we're going to get there further and faster if we do talk more together and help us all figure this out. Okay, let's move on to how do we keep pace with product innovation? Because EU AI Act is great, but I think all of all legislation itself, the concept of legislation itself, may itself be obsolete, because it kind of captures a snapshot and says, all right, it has to pass this bar. That's brilliant. Let's say all technology passes that bar now, but an update comes out next week, how do I know as a deployer of the software that the product stays in compliance?

49:36I don't really. All kinds of stuff is happening all the time. We might be coming to the point where products become living to a degree. They're constantly updating rather than waiting for monthly rollout. They're just changing all the time. So how does the legislation keep track of that and where where does the liability land is there any kind of uh uh grace if you like for someone who's bought pieces of software a year ago um thought oh well this is good and then suddenly a feature is released in it and it pulls you out of compliance any thoughts on how we handle that so sorry go on oh no i was gonna say so so you know a part of part of my role is to monitor regulation across the world and and watch this space next week for something interesting on that.

50:22But at the same time, I'm seeing a trend across the world when it comes to, you know, the good compliance around legislation and our technology. And so a lot of it is following a specific pattern. So if you are kind of working towards things like the UAI Act, then it will tend to cover most legislation across the world. However, to answer your point, you know, legislation doesn't get released, and you have to be compliant to it on the same day. And so you tend to have a grace period you know legislation will be live you know eua act we saw that you know august 2024 um you know so so we had plenty of time to get ourselves compliant and we've got even more time now so that's what we see with most legislation is you know legislation goes live it's passed by whatever you know authority is passing that um and then there is a an amount of time before um before it's live and you have to be compliant.

51:17So you do have plenty of time. But to go deeper on your point, if a vendor decides to create a whole new piece of their solution and a new product within their solution, it's also up to them to make sure that what they're bringing to market is compliant as well and gone through all the right checks. We work with vendors who do that pre-market deployment and make sure things are correct and compliant before they go out. The law bakes in some post-market monitoring requirements as well, and that comes back to what I mentioned about how important your vendor relationship will be if you're a deployer of the tech, because there should be a constant dialogue.

52:01If something gets updated, there'll be mandatory regulation required for paperwork that would sit around that and comms down to the deployer, the user. And similarly, there's obligations on those downstream users to report up issues with the way the tech is working, unintended consequences or changes that may be having an impact in a way that perhaps wasn't understood to be the case at the beginning when the tech was deployed. So post-market monitoring is already baked in. The provider, as Martin says, is responsible for updating and changing the way the tech might be used, deployed, or the unintended consequences.

52:37is they should be understanding what they could be, part of their good governance. And then there'll be a little bit of responsibility on you, the user of the technology, to say, this has gone awry. It's taking decisions that we didn't want it to, or it's putting emphasis on something that we never really thought was going to be the case. And actually, we need to solve this. So your vendor relationships are going to be very important. Yeah. And just to kind of make a point around this, You know, we talk about being compliant to the EU AI Act, and this is all kind of new AI based legislation. We're seeing, you know, dozens of pieces of legislation across the world, this kind of new world of AI.

53:16But actually, when we're looking now at court cases and administrative decisions that are going on, regulatory decisions that are going on, that's all on, dare I say, old law. So, you know, we've seen over the last couple of weeks Uber twice being brought to court over breaches of automated decision making under GDPR in the Netherlands. And we're seeing over in the US anti-discrimination law being brought against some vendors. We're seeing, you know, other laws around automated decision or privacy laws being brought up against vendors and employers over there as well. So actually, a lot of the lawsuits that we talk about in the industry, they're all on older laws and not on AI specific laws.

54:08So we have to make sure that we're compliant to them. And if we're compliant to them, most of the time we're in a good space to be compliant to the newer laws coming out as well. Yeah, very good. Folks, you have to go to the questions because there's a bunch of them here. Let's do that now. OK, this one from Sean Moran. I'm going to read out for the people who might be listening to this rather than watching it. So panel, is there a legal requirement if a candidate does not want to interact with AI, they can ask to engage with a human from the start of the process? So I'm applying for a job. I get invited to speak to Corey, the AI interviewer.

54:41I'm saying that I want to speak to a human being. Is there a legal requirement to provide a human if the candidate requests it? in a sort of roundabout way you kind of get to something not dissimilar to this but not not exactly under the gdpr so not under the euai act now the automated decision making rules um mean that you know if you are squarely within those rules if you're using technology that takes automated decisions which is quite a different quality of of legal threshold than what we have under euai act but if you fall into that then you can request human review or human intervention that's not quite the same as saying I want to substitute this for a human interaction but it it would sort of be pushing in that direction so no I think is the short answer but the long answer is you kind of get to the same result if you're engaged under automated decision rules under GDPR the EU IAC doesn't provide for that quite in the same way yeah yeah that'll do got it um let's move everyone thank you natalie we have sinéad here i think we've already dealt with this but let's try and deal with it again because i think it's quite interesting um okay why are employers reluctant to tell candidates that they use ai to score asynchronous interviews and will this be mandatory uh for eu-based employers after december 2027 and what about uk-based ones okay so firstly i think i'm not sure i'm not sure whether that's the case i mean um are employers reluctant to tell candidates they use AI to score asynchronous interviews?

56:13I don't think that's the case, is it? I don't know. Is it illegal? I mean, I don't know whether they're reluctant. I think that a lot of the organisations that I've worked with over the last several years haven't been reluctant to tell people. I think that there's a general kind of negative connotation around the use of AI within the recruitment process, and so that might cause some reluctance to shout about it, ultimately. But I've never worked with anyone who, when that's come up, has been reluctant to tell people that. Ultimately, as Natalie mentioned, you know, there's certain requirements in place now to make sure under things like GDPR as well and UK, you know, Data Use and Access Act to make sure that people are aware of it and make sure that they've got the opportunity to appeal against any decisions.

57:03and for those decisions to be reviewed by a reasonably responsible and experienced professional. But no, I've never seen anyone be reluctant about that. Yeah. Just to broaden that out slightly, I've seen it regularly on LinkedIn. I've seen people in recruiting responding to topics about, like, recruitment teams using AI to assess candidates and actively saying, recruiters don't use AI to assess candidates and I'm thinking why you're either lying or you don't go and you absolutely do yeah and I think I think there's some kind of dare I say kind of influencers out there that the kind of march along this track of you know there's no AI and ATSs I've used all the ATSs in the market and you know what have you and based on based on the number of people number of vendors and AI screening vendors that I've worked with that's not the case.

58:03But I just think that there's two sides of, you know, unfortunately, there's some influencers out there that may be not getting it too right. No, very good point. And it's, it's impossible not to use AI these days. I mean, you switch on any, any app, it's going to be AI in some way. It's not something it's, it's, it's basically a conversation we need to move away from, I think. Okay, Ezra, I can't get to your questions, because we're running out of time. I I want to have a final question here for Martin and Natalie, which is, OK, what's the how to as it stands today? Most people in the audience are actually operators.

58:39They're recruiters. Right. What do they need to do, if anything? Natalie. I think not be scared to implement technology that is genuinely useful and creates value. But think of it like that car analogy. It needs to be done safely. get the right people involved, start having open conversations about how you can deploy in a way that's compliant, pick really good quality vendors that can support your compliance. You know, don't skip that step, but don't be scared to do it. Yeah, fantastic. And also, by the way, this opportunity for recruiters to expand scope, you know, if you can be like an expert on the legislation and how AI sort of shows up in your business, that's going to be at value in your business.

59:25So this is something that your C level might actually be looking for. Martin, your advice, please. So we've covered some things that are already enacted and in place and live. So we should be investing in supporting AI literacy. We should be making sure that any kind of interaction with AI systems is transparent and people know about it. But in terms of getting ready for the AI Act, first thing to do is audit your stack. You'll be surprised what you might be using that you don't realise you're using as well. So make sure you're auditing your tech stack. Make sure you know what it is and what it isn't doing.

1:00:02And the other thing is to start thinking about human oversight. We're seeing this massive trend going on across both sides of the Atlantic right now around what is human in the loop, what is human oversight. And it's not just rubber stamping the output from an AI system. It's making sure that there is true human oversight in the system and that you are looking over every single outcome that comes out there. Yeah, brilliant stuff. Folks, let me know in the comments if you want us to do a round two on this. I personally think we should, but I think a quarterly rhythm on this topic is probably quite important for us to just touch base on and keep addressing where we're at.

1:00:39Okay, that's about it. We're out of time. Thank you, everyone, for joining the show. Adam Gordon, Natalie Farmer, Martin Redstone Alexandra Daboli, thank you so much for trying your best to get on we'll have to fix that next time I'll drop you a line sort of offline after this call and yeah thanks everyone for watching folks, I hope you enjoyed it we'll be back next week as ever we're going to be talking about point solutions versus platforms, it's that story again, disconnected AI agents versus sort of ecosystems with multi-agents in it which is the direction of travel. How do you build a tech stack in the agentic era?

1:01:16Okay, September is also set up. So if you're interested in all of the program in September, follow the channel on Crowdcast. You get updated all the way through there. If you're watching this on LinkedIn, just follow my LinkedIn profile. Basically, all the shows go out on LinkedIn, my profile as well. So if you don't want to sign up to Crowdcast, you can just watch it there. All right, that's it, everyone. Have a good week. I'll see you next week. Bye-bye. Thank you. Bye-bye. Natalie, Martin. Can I just ask one, the show's over.