IBM Think is an email newsletter from IBM covering AI, cloud, security, and sustainability industry news, events, and insights. It is distributed by subscription; the video promotes it as a source of monthly AI updates, while IBM's registration page describes the subscription as weekly.
Security Intelligence is IBM’s weekly cybersecurity podcast. Its episodes combine news roundups, expert panels, and in-depth conversations with cybersecurity practitioners, including IBM X-Force experts. The series covers malware, attack trends, vulnerabilities, defensive tools, and emerging security issues; monthly bonus episodes feature longer interviews about practitioners’ work.
Searchable transcript of The PHP bug threatening the web. Plus, who's auditing AI? — IBM Technology (20:22). Search for a phrase, then click its timestamp to jump straight to that moment in the video.
Captions sourced from the original video on YouTube, published by IBM Technology. The video, its captions and all related intellectual property remain the property of their respective owners; AINotes claims no ownership. Provided for research, accessibility and search — see the Transcript Notice and Copyright Policy.
00:00 Panelists, I've got one question for you today. What website would leave you devastated if it went down today? >> It's got to be X. I need my doom scroll. >> I think for me it would have to be um Google. So, how else am I going to look everything up? >> I think the internet should be out at least three or four times a day just so people can go read a book or go outside and look at the birds and listen to the trees and the wind blow.
00:20 That's what I think. >> Jim, that was the best answer, man. I could concur. Hello and welcome to Security Intelligence, IBM's weekly cyber security podcast where our experts turn the biggest industry news stories into practical takeaways that you can use. Our panelists today are Nick Bradley, manager of XForce Threat Intelligence, Don Cohen, X Force Red Associate Hacker, and Jim Gwyn, America's cyber security leader.
00:51 And we've got a variety of stories today from the critical to the concerning which I guess is also kind of critical. We're talking PHP vulnerabilities affecting the fabric of the web. We're talking enthropic embedding evaluators in its own company. And we're talking not so anonymous security issues with a very popular anonymous networking service. We found some pretty severe bugs in in PHP.
01:18 I don't know about you guys, but when I saw the news, uh sort of uh Star Wars crawl sort of appeared in my in my head, you know, PHP compromised. The vulnerability itself is dubbed PHP to shell. Um the PHP Foundation has released patches addressing these two vulnerabilities as well as a statement on how security updates will be handled in the future.
01:40 I think this is a huge news story first of all that PHP is so ingrained in the in the web and that these um vulnerabilities would allow for a massive sort of exfiltration of user information and data. The first thing we have to understand is the the severity of what we're dealing with here and why, right? Because we get vulnerabilities of this nature all the time, right?
02:04 It's vulnerabilities are discovered, they are ethically disclosed, they are patched, and usually you don't even hear about them until the patch is released. The issue we have here is this time things didn't go according to plan, right? the uh the vulnerabilities and information on how to exploit them started leaking before the patches were available and that's that's of course that's a recipe for disaster.
02:29 So yes, these were discovered through AI. Uh, lucky for us, the exploit chain is difficult to take advantage of, but what it really shows us is how vulnerable we are when we have one single point of failure across 70% or more of the web. >> We definitely agree with you there. Um, I'm wondering in terms of the patching process, how in in your experience does a team find a vulnerability and um sort of collaborate with one another to address that vulnerability without compromising the rest of the of the service?
03:08 >> The first issue is you have to know where it's at. You don't know you have a problem if you don't even know where it's at. And that that is the issue is PHP is so embedded and in so many different things that might not even be apparent that you have it. And then the fact that only uh the 8.1 and prior are all vulnerable and will not get patched is another problem because that again that's not going to be patched and you have so many systems that are still using those that legacy software and it's got legacy system
03:39 upon legacy system. So if you do upgrade it to another version which is required, you break legacy systems that are running behind it. So this definitely takes a strategic approach in order to solve the problem. Patch management isn't the only issue. Of course that needs to be you need to have a a seriously agile patch management program here and you need to be willing to bring things down and suffer downtime.
04:04 But you got to know where it's at before you can fix it. I'm noticing just going off of what what Nick just said, the downtime aspect of of um of all the software, you know, when it comes to updating firmware and releasing patches, downtime. And you know, that seems to be a huge a huge factor when it comes to making time to deploy these firmware updates and things like that.
04:26 How can a team sort of bake that into its standard operating procedures so they know that when a when a critical vulnerability is discovered or a new firmware update is required, they can say, "We've got a plan in place. We can make this happen ASAP with minimal downtime, with minimal sort of interruptions." Should that be part of the plan or is that something that you know you may have to take, you know, as it as it comes?
04:50 every time there's a black swan incident and it doesn't really matter what it could be want to cry, one of crypto, it could be there's there's there have been hundreds over the past decade that people know about and they're and they're real events and they have significant impact. The bigger question, my opinion, the bigger question is not um what can we do to mitigate this moving.
05:12 It's more about how do these organizations, even open source organizations or for-profit companies that build and and manufacture software and and and hardware with firmware embedded, how do they use the tools that these researchers and ultimately these attackers are going to use to better make their products more resilient? That to me is the really big question.
05:33 How are you going to take like soul 56 which is what was used by these researchers to be able to figure this out and then whiz and others figured it out or not figured it out but check their math if you will. How are you going to use those same tools embedded in your organization on a day-to-day basis to be able to test and secure to the best of your ability the highest level of cyber resiliency that you can?
05:54 That to me is really the the bigger question. And when are organizations going to start to truly leverage the secure by design um uh uh methods and and procedures embedded into their software life cycle because right now today there is a there in today yesterday and I don't know three five seven years ago there's been a propensity to try to get the latest technology out and the latest software packages out and we might have missed some uh some vulnerabilities and now we have the capability with these frontier models to
06:26 be able to test, look for and build exploits and then remediate them before we ever release a product. So I think that's really from my perspective looking after you know the portfolio that I have a chance to look after. I I think that's really the major question. There's going to be another black swan. This is not the biggest event that's going to happen in the next I would say 18 to 24 months.
06:46 There's going to be bigger ones, but it's how do we use the technologies that are discovering these vulnerabilities that bad actors are going to use to exploit these vulnerabilities. How do we use them to make us better and more cyber resilient as an ongoing part of our our our secure by design uh strategy? You >> know, Don, building off what what Jim said, you know, it's important to understand the tools you're using.
07:08 Um and in PHP's in the PHP foundation's case um you know these vulnerabilities were discovered after the foundation underwent its first external security audit in a few years which I found very fascinating and a little bit um concerning that it had been um a while since they'd undergone a security audit like this. What is the benefit of undergoing an external security audit?
07:35 What can a company sort of expect there? So one of the biggest things is it challenges the assumptions that you have as an or you have as an organization. What what are some of the security assumptions that you have made or your team's made that you that you think are you have things that mitigate uh vulnerabilities or protective measures in place um until you truly test them from an outside perspective, you're not necessarily going to know for sure if they're actually effective.
07:59 Um just because it's it's just human nature to think, okay, I've already covered that. um I'm just gonna move I'm going to move on to the next next thing I have on my to-do list. So having an external organization come in and give you a give you a good an outside view that's not biased will uh do do pay dividends. >> Yeah, Don, a really good comparison to what you're talking about there is the reason why you have someone else proofread something you write, right?
08:30 because you are seeing what you want to see and someone else is going to look at it completely differently. So that's I think it's a great thing to do and it should be done more often than not. >> Nick, you should be a uh a book editor. >> Funny, I spend a lot of time doing that. So good call. >> Our next story involves a little bit of uh someone looking over your shoulder.
08:57 Um if that someone was a team of six researchers and you and your shoulder were a giant AI company. Uh Anthropic has invited a few meter uh researchers that is model evaluation and threat research. Um the this is a company that investigates um AI models. um has invited these researchers onto its campus um outfitting them with office space, company hardware, and quote, I found this very amusing, employee like access to internal tools and systems that are roughly on par with what their own risk assessment teams have.
09:30 Nick, I'm just I'm just going to cut to you very quickly. Um I saw the look on your face. um you don't really seem too happy with um getting employee like access when your job is um investigating and evaluating uh frontier models. >> It sounds like mental gymnastics. It just you you just hired a bunch of contractors to come in and do the job, which is great.
09:51 I have no problem with it. I just don't understand why the why the misleading cloudy speak to call it what it is. >> Yeah, it's very true. There they are. Meter is one of a group of u many evaluators that Anthropic is is teaming up with essentially to investigate and evaluate its models. Um these these evaluators are are divided into almost tiers of access which is which is um which is what makes the meter employee like access quite interesting to me as the meter arrangement is the highest tier of access according to
10:26 anthropic. third tier of access which includes um publication rights um to what it has what it is researching when it comes to um anthropic. The first tier um involves a quote narrow scope investigations um that allows evaluators to publish with approval of anthropic. The second tier with the second tier evaluators can publish without approval. But Anthropic reserves the right to restrict what um information um is published um but allowing the second tier evaluators to write sort of rebuttals on why the redactions um
11:01 may be unnecessary or unwarranted. Jim, I'm going to throw it to you. How does that um how does that correlate with anthropic sort of stated goals of pacing AI development and regulating AI making it safer for businesses and consumers to use these companies right it could be you know it could be open AI it could be anthropic it could be Grock it could be you know uh Google they they all have immense capabilities with their with their models to do some very very powerful things sometimes they're good hopefully most of
11:34 the time and sometimes they're not So good in in cases that we've seen there are models in the world right if you take the big four right Accenture EY excuse me EY PWC KPMG and Deote they do what's called peer review and they go in and they selectively audit each other and they look at their each other's audit papers from public audits in the past to look for things that maybe the primary auditor missed.
11:59 So, you know, I don't know what the rotation is, but you know, PWC will look at EY's audit papers. EY will look at at Deoites, Deoites will look at KPGS, and they rotate, and they rotate around, and they do that so that they can test each other's um audit methodologies and their ability to discover things that uh maybe a company uh mistakenly did or or uh intentionally tried to hide.
12:21 I think that there's a bit of a um uh um I think that all of these frontier companies, all of the big ones, they could take a play out of that playbook because that's been highly effective in discovering u uh mistakes and vulnerability and audits. And I think it can be the same way with the technology. So if you took all the leading frontier companies and they did a rotation where they looked at each other's models and through responsible disclosure internally and then externally shared the results, I think we as
12:53 public would feel a heck of a lot more comfortable that they're doing the best that they can to create the most secure models that they that they're that they're able to do because their peers are looking at it. It's not somebody that's under contract. It's not somebody from a third party. There's not this taring system that goes on. It is literally your biggest competitor is looking under the hood and looking for vulnerabilities that they might find that you didn't discover.
13:16 Don, you mentioned this a minute ago, right? It's like, and Nick, you said, you know, it's like, you know, reviewing someone's um their thesis or their book or reviewing their paper or peer reviewing. That's what this is. And there is a very very structured model for this in all public companies with the biggest auditors in the world doing that same sort of peer review.
13:36 I think that's what these companies should really pay attention to and try to do because if they did, I would feel a lot more comfortable that the researchers from one of the leading frontier companies is looking at another leading frontier company and discovering vulnerabilities that they didn't know existed and then sharing with them to make it stronger and better and make it more resilient moving forward.
13:56 I like Jim's assessment there because this it's it's what they're doing seems like something that was already there and the reinventing of the wheel whereas this really just feels like an overblown complicated publicity stunt to say look at us and what we're doing and I I mean I I don't have anything against anthropic and that's just my personal opinion though I just I feel like this was uh overdone.
14:17 I think that there's a better way to do it and I know that there is uh there is an industry that is you know over a hundred years old a century old that has been doing it a specific way and we have a lot of asurances and every single public company in the world has this rotation of auditors and the auditors audit each other's work I think that there's something to be learned from that and it would make the world feel a lot a lot more comfortable that these uh frontier companies these frontier AI companies are doing the
14:46 best that they can to have the most resilient models that they can produce. >> It's interesting the the financial institution analogy is very apt. Anthropic uses it um in its own sort of explanation of of why it is um bringing onboarding these these evaluators. Don I I I want to ask you maybe what your what the dream scenario is when it comes to embedding an evaluator in a company like this.
15:12 like what would you like to see come out of this? Um, you know, as a as a sort of as a security researcher, like is there anything about this that that excites you or that gives you pause? >> So, it's actually it's pretty interesting that they're that they are doing this. I agree with both these other gentlemen that that more than likely is a publicity stunt, but it would be really interesting for to be able to see some sort of to see what work they can do um using that outside perspective and what vulnerabilities they
15:41 can find. Now, it it would would be a good proof of concept if they were if they were able to to show like good results um to kind of make it make it more more than a a publicity stunt by giving delivering the goods. So, I think that would be the ultimately the best outcome for for that program, but it's it is also something you can you can accomplish by bringing in outside individuals as well.
16:10 Our final story has a few interesting layers to it. That'll be funny in in a minute, don't worry. Um, for a few days in October, um, 333 tour exit nodes were hijacked with anonymous traffic and emails redirected and intercepted by an unknown actor. Um, if you're unfamiliar, tour, the onion router um is an anonymous um networking layer that allows people to um communicate and transfer information over the internet anonymously.
16:42 Um very useful when you want to hide things like your IP address and your location, encrypt your communication like emails and stuff, things like that. which is why um the compromised exit notes are a pretty concerning um issue when it comes to when it comes to the tour network. Um according to AR Technica and tour volunteers, an attacker reused valid passwords to access a tour management server and from there installed a Linux backdoor um and credential stealer.
17:14 Nick, I might be a little younger than you, but you know, grow grow growing up, you know, tour to me had always sort of been synonymous with secure anonymous, you know, network connectivity. I wonder if that's the same in your circles, and I wonder if anything has changed maybe from my time uh in college using Torah and and now. >> That's just an assumption, especially as a as a security professional.
17:36 I'm not saying it doesn't have its place in its use and it's I mean anonymity is it's got its place but I have always felt like to anything tour you need to just assume it's bad. I I agree with Nick um wholeheartedly. Um when uh if you if you stay in this profession long enough, you're probably going to do things uh that require you to have uh to be anonymous, to have anonymity, and you have whether it's research, whether it's hunting down the bad guys, and you have to use the tools that are available to do that.
18:10 >> I've just assumed that every tour exit node is compromised from the get-go. So there you go. >> This is the sort of overarching consensus. Don, I'm going to go to you. Maybe you can uh join me over here on the tour good side, but I'm I my don't have high hopes for that. >> Unfortunately, no. Like I have to agree with I have to agree that generally the majority of the things you're that you're going to see like from on the tour network are not going to are not going to be good.
18:32 Otherwise, they would it's because it's much easier to host something on the public internet and to get people to go to it than it is on the tour network. Um, but it does have I would say it has benefits for certain types of like individuals like say for instance individuals that are trying to circumvent censorship so on so forth. While that is illegal in their country, it might be some a goal they have.
18:52 Um, so I wouldn't say it's necessarily immediately nefarious, right? That's just that's an example I can think of. You know, one one interesting point I will I will note um and maybe Nick you can you can illuminate this or or something. You know the RSA key used that um researchers could not verify or or or authenticate along with the reused password.
19:19 Is this a classic cyber security hygiene flub or is this a larger issue when it comes to um you know browsing anonymously? >> No no no it's just a flub uh in my opinion. Okay. I can't prove it, but I'm just going to say that in my opinion, it's just a flub. And the reason I say that is because we've already seen over this year alone, the amount of threat actor websites that were compromised, whether it be by, you know, um, legal investigative agencies or other threat actors, all due to just bad hygiene on their side as
19:50 well. It just shows that even being a threat actor doesn't make you immune to bad security practices. >> That does it for our for our episode today. Thank you to our panelists, uh, Nick, thank you Don, thank you Jim. Um, and thanks to our producers, um, and thanks to you for watching and listening. Subscribe to Security Intelligence wherever podcasts are found so that you never miss an episode. And stay safe out there.