37 minSeason 4, episode 4

The Invisible Code Problem: When You Can’t See the Attack, Can You Stop It?

With Chris Romeo, Izar Tarandach, and Matt Coles

Software supply chain & open source

In this episode, we dive into the strange world of invisible Unicode attacks and what they could mean for modern software security. We explore how hidden characters can be used to conceal malicious code within packages, why this isn’t entirely a new problem, and whether current tools, such as linters and SAST, are equipped to detect it.

Episode title image: The Invisible Code Problem: When You Can’t See the Attack, Can You Stop It?

Listen to the conversation

About this episode

In this episode, we dive into the strange world of invisible Unicode attacks and what they could mean for modern software security. We explore how hidden characters can be used to conceal malicious code within packages, why this isn’t entirely a new problem, and whether current tools, such as linters and SAST, are equipped to detect it. We also question the role of LLMs in both enabling and detecting these attacks, and whether this is a real emerging threat or just another overhyped security scare. Per usual, the conversation is filled with sarcasm, skepticism, and a healthy dose of “just don’t do it.


Follow AI Security Table:

➜ Home: https://securitytable.ai/

➜ X: https://x.com/AISecurityTable

➜ LinkedIn: https://www.linkedin.com/company/ai-security-table/

➜ YouTube: https://www.youtube.com/@AISecurityTable

Transcript · 37 min conversation

0:13Chris Romeo Hey folks, welcome to another episode of the Security Table where Matt can't stop laughing. It's been a while since you've heard from us. We were traveling the world. No, we weren't.

0:27Matt Coles In our minds.

0:28Chris Romeo I wish we could say that.

0:29Izar Tarandach In our VR headsets.

0:30Chris Romeo In our VR headsets. We were at the Eiffel Tower.

0:33Matt Coles We were at the Fear of the Mind.

0:37Chris Romeo Yeah, I don't want to wait. I don't even know what that means. The Fear of the Mind. Hold on. Now, Izar, if you're watching the video stream, appears to be striking a pose. He's about to say Vogue for all you kids of the '80s. So yeah, I'm Chris Romeo, joined as always by Matt Coles and Izar Terrindosh. and we unpack the wild, the crazy, the, uh, underside of security. What is on the underside of the security table? Is it just gum? Like used chewing gum or what would we find there?

1:16Izar Tarandach That's an interesting question. I never considered that one.

1:19Chris Romeo Don't you dare ask an AI agent either. Get outta here with your AI.

1:23Matt Coles You're asking, who are you asking there, Chris?

1:26Chris Romeo What's on the underside of the security table? Is it just, did someone just chew gum and like stick it under the security table?

1:32Matt Coles To see the underbelly of the security table.

1:35Chris Romeo Ooh, security table after dark.

1:39Matt Coles That's like a completely different, different episode.

1:41Chris Romeo Yeah, that's true. All right, well, let's get into this. We got a couple of, uh, let's deal with this. A couple things to deal with here. Let's deal with this supply chain attack one first, cause I think it's the easiest one. Because, uh, what is old is new again, I think is the answer. Because you guys remember the first Unicode attack, right? You're old enough for that.

2:04Matt Coles Oh my God. Unicode is, was, was, maybe still is horrendous. Is it double? Is it single byte? Is it double byte?

2:13Izar Tarandach No.

2:14Matt Coles Is it triple byte?

2:15Izar Tarandach It's 4 bytes.

2:17Chris Romeo There's bytes for everybody. The legacy attack. If I remember, this is, we're going back a while, but I believe that the legacy Unicode attack was was equivalent to a dot dot slash. Well, at the end of the day, let you escape directories and things and execute code in places. I think there were, there were a couple of things, right?

2:36Matt Coles There was, first off, there was the, the, the stack overflows, right? So you had Unicode characters. If you had, if you only planned for UTF-8 and then they suddenly got UTF-16, you got double byte characters and you can overflow buffers.

2:49Chris Romeo Okay.

2:50Matt Coles Right. And then you also had, um, like similar looking characters, characters from a different character set that allowed for, um, you know, phishing and spoofing and whatnot. Um, and then I guess you were talking about characters that behaved equivalently as./, um, wasn't familiar with that one, but, uh, that one sounds possible.

3:14Chris Romeo That's the one that I saw in the incident response world. That's the one that, uh, our customers were getting hit with was the—

3:21Matt Coles well, that was, you weren't thinking about Zip Slip, were you? Cause Zip Slip knows where you have embedded dot dot slashes in zip files later.

3:27Chris Romeo Yeah, zip slip was later. I feel like the Unicode attacks were years before the zip slip attack, but it's not entirely possible.

3:35Matt Coles Yeah, I mean, Unicode is like IPv6, right? It's, it's been around for ages, and people— some— now people use it all the time without— probably without thinking about it. Um, although if you're in the Python world, you have to think about it constantly because I don't know about you, but I get character encoding issues to on occasion.

3:55Chris Romeo That sounds like a reason to use a different programming language.

4:00Izar Tarandach No comment.

4:05Matt Coles Show me one that handles it properly every time.

4:08Chris Romeo Perl! Hold on, let me just have AI create one real quick.

4:13Izar Tarandach Yeah, but the problem in this one, bringing us back to the present, is that apparently there is this new thing out there where UTF is being used to add invisible code inside packages out there.

4:35Matt Coles And, uh, and of course humans can't see invisible code unless you have x-ray vision or, you know, see invisibility or something.

4:45Izar Tarandach Which to tell you the truth, I think it was a wasted name of an attack. This could easily, they called it Glassworm.

4:52Matt Coles It could have, wait, they named it?

4:53Izar Tarandach Yeah. It could easily have been the attack of the invisible characters.

4:59Chris Romeo Glassworm. Glassworm.

5:00Matt Coles Who calls it Glassworm? That's lame.

5:02Chris Romeo Right.

5:04Matt Coles And the invisible code.

5:07Izar Tarandach Yeah.

5:08Matt Coles The shadow knows.

5:10Izar Tarandach They, they say, okay, you, you get this invisible code inside popular packages that get downloaded, so it becomes a supply chain issue. And SAST, conventional SAST, won't see this because the characters are quote unquote invisible.

5:27Matt Coles So, well, it, the SAST will still see the characters.

5:32Izar Tarandach It just won't interpret them as anything nefarious. Which brings me to, to, to my first point here. They gave an excerpt of code found out there. And it's, uh, as expected, a loop that's mapping the, uh, UTF characters into something that's more readable and blah, blah, blah, and filter and whatnot. But then comes the last line, eval buffer from string. So, uh, looking at my, looking at my watch, it's 2026. I was still looking at code and seeing eval and not saying this is evil. I mean, talk about unintended consequences, right? Fine. You have this whole invisible thing happening and it's nefarious and it's horrible, but it ends up in an eval.

6:26Chris Romeo Yeah.

6:26Matt Coles Your, your linter should catch that. Never mind static analysis.

6:30Izar Tarandach Your freaking eyes should get that. But the problem is that at, at the end of the, uh, What really irks me, the last character of the article, and these are not the original researchers, it's just somebody writing on the research. The best way to protect against the scourge, scourge, it means that it's all over the place, right? Thousands and thousands.

6:55Chris Romeo That's an LLM word.

6:57Matt Coles It could be anywhere. It could be anywhere. Who knows?

7:00Izar Tarandach Right. On the other side of the security table, including. Um, of supply chain attacks is to carefully inspect packages and their dependencies before incorporating them into projects. This includes scrutinizing package names and searching for typos. Yeah, let me go to my developers and just tell them that now I expect them to check for typos inside the code.

7:23Matt Coles Wait, wait, wait. If, if everything's invisible, what are they looking for?

7:30Izar Tarandach Clearly for the evidence of absence of evidence of, uh, knowns, unknowns, plus typos. Like, look for something strange.

7:41Matt Coles Well, why typos though? I mean, if you, if you can, if you can inject invisible characters into code, why can't you inject invisible characters into a title and make it properly named with no typos with invisible characters? With embedded characters.

7:59Izar Tarandach Hey, therefore, that's why they continue the paragraph and gave you some clarity. If suspicions about LLM use are correct, malicious packages may increasingly appear to be legitimate, particularly when invisible Unicode characters are encoding malicious payloads.

8:18Matt Coles So, again, what is the, what is the, what is the, what is the mitigation here? It's for humans to look at something that they can't see.

8:26Chris Romeo Yeah, that sounds like a plan.

8:30Izar Tarandach No, no, no, no, no. Wait, not only they can see, but also has typos.

8:34Matt Coles How about, how about a simple, how about a simple grep looking for a certain set of Unicode characters? I mean, add it to your linter and be done with it. Right. If you find it, you don't need to review any further. It's suspect. Throw it away. Right. They did say that there are legitimate cases for this. I don't know of any. If you, if anyone has any good examples of when you would want to use UTF Unicode characters that are invisible to the, to humans, maybe for control codes, you know, maybe it's control characters we're talking about here, but if I want to store a secret in my code, uh, would you store it as invisible characters?

9:12Chris Romeo Yeah. I was just joking.

9:13Matt Coles For those who don't know, for those who aren't familiar with that term, security by obscurity, or in this case, security by invisibility.

9:26Chris Romeo I mean, that's cool.

9:27Matt Coles This is just, I don't know, this, and this is not an LLM problem, right? This could have, this could have happened for decades.

9:34Izar Tarandach Oh, there's more.

9:36Chris Romeo Oh, there's wait, but wait, there's more.

9:39Izar Tarandach Yeah. Let's make it an LLM problem because here in the middle, uh, the researchers suspect that Glassworm. The name they assigned to the attack group, is using LLMs to— no, that's not the one. Sorry. Sorry. Um, oh, the invisible Unicode characters were devised decades ago and then largely forgotten, as they should have been, or actually as they shouldn't have been devised. That is until 2024, when hackers began using the characters to consume malicious prompts fed to AI engines. While the text was invisible to humans and text scanners, LLMs had little trouble reading them and following the malicious instructions they conveyed.

10:24Matt Coles So wait, it's not white text on white background anymore?

10:26Izar Tarandach Yeah, but you know, I, I, I'm a bit confused here because even then, if you're using this, this invisible characters, it's not an A, B, C that's invisible. It's a bunch of characters that are not represented. Like visually, but there's a tail between us.

10:50Chris Romeo A tail of two cities. It was the best of times, it was the worst of times.

10:55Izar Tarandach That's Nessie, the cat from Loch Ness. At least it's not Sir. Okay, if you're not watching this, Matt has just been invaded by a tail of a cat. Okay, cat's gone.

11:09Matt Coles He'll be back, I'm sure.

11:12Izar Tarandach He'll be back. So how does the LLM actually interpret these Unicode characters as a bad prompt if their only meaning is being in a range that's not visually represented? I'm confused here.

11:27Matt Coles Well, also, I mean, we know how LLMs tokenize. We know how, we know that LLMs tokenize strings. And so does the tokenization actually catch the individual characters enough to convey meaning. I guess, you know, an interesting experiment, you could ask an LLM to produce sentences with invisible Unicode characters. It knows the entire character set, right? And it should be able to identify those things. Like, show me, you know, tell me the book tell me which books have this string in it.

12:13Izar Tarandach Now you're using an LLM as a grep engine to look for.

12:20Matt Coles Show me all the code package. I'm looking for a piece of code because I want to use it. Show me all the code packages that have this construct.

12:29Izar Tarandach I don't think that that would work.

12:32Matt Coles I don't think it will work well. Cause it may just invent them.

12:38Izar Tarandach It probably would, but anyway, it would try and use a tool to go and search the web. It wouldn't look inside its own model and see what representations do I have in here.

12:48Matt Coles What if you told it not to?

12:50Izar Tarandach Hmm?

12:51Matt Coles What if you told it not to?

12:54Izar Tarandach Still, it would probably just invent something because I don't think that that's how the query bits works. You can't, I don't know.

13:05Matt Coles It can predict, remember it's a prediction engine, right? So it won't do it reliably.

13:12Izar Tarandach That goes without saying it won't do anything reliably.

13:16Matt Coles Or repeatably, maybe reliably is, uh, it's probably the wrong word. Repeatably is the—

13:21Izar Tarandach you can't rely on the, on the consistency of the result, right? It will do things reliably, but what you get out of it is not reliable.

13:30Matt Coles Someday we should have a talk. We should have a chat about rule-based versus LLM-based processing, given the non-determinism and how that affects security.

13:40Izar Tarandach Well, we could just bring that down to LLM-based threat modeling and rule-based threat modeling. And, you know, one of them is always going to give you the same result given the same input. The other one, your threats may vary, but it would be interesting to me to know how much constraint, how many constraints and how effective they would be if you would try and make the LLM-based one act as if it would always give you the same thing.

14:17Matt Coles Isn't that what RAG is for? Confirm, validate your results against this dataset. And only use this dataset.

14:23Izar Tarandach No, that, that's, that's for, that's for, uh, grounding the, the, the, the, the, like the space that you're going to answer about. It's saying don't go around and invent stuff or go look for stuff.

14:37Matt Coles But, uh, my problem is that you constrain this dataset, right?

14:42Izar Tarandach But that doesn't guarantee that the result and resulting threads that you're going to get are going to always be A, B, C. You could get C, B, A. You could get A, G, and B.

14:52Matt Coles Right.

14:52Izar Tarandach That, that's, that's where I would be into. I'm sure that some university out there, probably the guys in Belgium have already looked at this kind of stuff. And actually I'm going to look for it as soon as we finish here. But, uh, again, those guys do the great work. Huge, huge respect.

15:12Matt Coles Who needs invisible characters when you can just, uh, pose a different question to Isar and Sidetracked.

15:18Izar Tarandach It's shiny. It's shiny.

15:20Chris Romeo So what do we, at the end of the day, what do we do with this? Is it just a new class of attack that we got to worry about?

15:27Matt Coles Does it? It's a simple fix. Again, just add it to your linter looking for those Unicode characters. If you find—

15:32Chris Romeo but what are you looking for? What are you scanning for? The, cause these are packages. So scanning the packages, you're scanning the packages for a range of characters.

15:41Izar Tarandach Uh, I'm, I'm sorry, but I'm going to go back to the roots here. Like at some point you have a string that maps to whatever is bad and that string has to be executed.

15:52Matt Coles Well, yeah, but if it doesn't, if it goes to eval, that's obvious, but what happens if it goes to a Popen or a system that you expect to see, right, in the code normally?

16:04Izar Tarandach Would you expect to see something that jumps into user-provided or hardcoded stuff and just starts executing it.

16:14Matt Coles It doesn't have to necessarily be that if you take an input stream and then you tack on characters to it, right? Maybe you get a command injection. Maybe you introduce a command injection that wasn't there previously, but again, or a SQL injection, but something is going to be executing potentially tainted code.

16:33Izar Tarandach You can flag for that.

16:35Chris Romeo This isn't, this ultimately a repo problem. Like, shouldn't the repo just block the upload of a new version of a package that has invisible code in it and call it done?

16:46Izar Tarandach That's one way, of course, but not everybody takes code from repos.

16:51Matt Coles But also, well, I think what, I think what is, uh, what is, or what, uh, Chris is saying is if somebody is going to maliciously inject code into a repo, you can stop it at that point and don't allow it to be introduced right in the PR. For instance, I mean, that doesn't stop a legitimate developer, like a maintainer, from introducing code because they might bypass the PR process.

17:16Chris Romeo Yeah. But if you had some type of check, like, I believe that when I upload a new version of a package to a repo, isn't there some scanning that's happening to that package?

17:29Izar Tarandach Some of them. And. Of course, as we well know, different kinds and levels of scanning. But I think that these guys are saying that SAST usually doesn't see it. Yeah. Static analysis tools see only whitespace or blank lines.

17:46Chris Romeo So the scanner might not even—

17:48Matt Coles wait, whoa, whoa, whoa, whoa. Hold on. The, the SAST tools are not looking at what gets rendered. The SAST tools are looking at what's, what bits exist.

17:57Izar Tarandach Exactly.

17:58Matt Coles And so as long as you're not encode— not, they're not being doubly encoded in a way that the scanner can't unpack them. Right. So scanners won't go into Base64 encoded because usually that's, those are strings that it doesn't need to look at. If that's a code block, if that's an executable code block, then that's a problem, right? The scanner should be fixed for that. Um, again, a linter. right? And a linter, you could then look for variations or variants, right? And forget— so linters are just basically grep, right? They're basically looking for string patterns, regexes. They're not— I'm not saying any of these are universally perfect, but they get you to a place where at least maybe you flag suspicious things, right?

18:48Chris Romeo So we're basically doomed. We're doomed.

18:50Matt Coles Or do you know?

18:51Izar Tarandach Because, because the next, the next line, it says to a Java, JavaScript interpreter, the code points translate into executable code. What's, what's ringing in my head is that you can't have it both ways. You can't have SAST ignoring it as whitespace and JavaScript looking at it and seeing executable code.

19:09Matt Coles Yeah.

19:10Izar Tarandach Because you're missing a mapping in here and that mapping is going to get a string transformed into something that can actually be. Interpreted. That's the point of an interpreter, right? So that's an interesting—

19:22Matt Coles actually, and that's an interesting question. So they're using, they're using JavaScript, which is an interpreted language, right? How does, how does this multi-byte Unicode get compiled if you mix character sets in as executable, as compilable and linkable executable code? Does the compiler actually handle that properly?

19:43Izar Tarandach Look, it's beyond that. No, it's beyond that. It's beyond that.

19:47Matt Coles Because, you know, I mean, if you try this in C or you do, I mean, it may, it may not work, period.

19:53Izar Tarandach No, it doesn't matter. It doesn't matter because whatever you create that is not part of the language syntax is going to get flagged by whatever is compiling or interpreting it because it's not part of the language syntax. So in order to use those things, you have to put them into some block. That is not compiled or interpreted. So it's either a comment or a string. If it is a comment, compilers will drop comments unless you explicitly tell them, and I don't even remember if that's possible nowadays.

20:25Matt Coles If you, uh, if you don't strip, well, you can, yeah, uh, compiled languages, I think, will always strip comments.

20:30Izar Tarandach Yeah.

20:31Matt Coles Right. It's symbols and things that you, that you may have.

20:33Izar Tarandach And strings, I'm sorry, the language is just doing what the language is supposed to do and it doesn't care about what's inside the strings. It's just bytes for it.

20:41Matt Coles It just puts it in a data block, right?

20:43Izar Tarandach It doesn't matter if it's multi-byte or whatever, that's in the construct of the, in the format of the bytes, not in the fact that there are bytes in there.

20:51Matt Coles Right. And so then you have a, then you have a reader for those bytes that would be, have to be multi-byte aware.

20:57Izar Tarandach Exactly. But then it brings back to my initial point. You had a mass of bytes in there. It didn't get compiled. It didn't get interpreted. It got remapped. Into something that can be executed. And then you have some jump, some entry point, something that makes that thing be executed. And that's where you can flag the whole thing. So unless you have a package that, as you said, expects to see evals of strings, in which case you shouldn't be using that package because frankly, that's stupid.

21:35Matt Coles So never, never mind any control codes whatsoever.

21:39Izar Tarandach No, I mean, you know, I'm, I'm just, it looks like a very interesting academic exercise. It's almost like, I don't know, can we call this steganography for executables? But beyond that, I wonder, tools should be there to, to catch it.

21:59Matt Coles I wonder though, if there is a legitimate use for this, thinking about, if you think about things like memory-constrained code, right? You're looking at embedded code or firmware. Is it memory efficient for these Unicode characters to exist? They're probably multi-byte characters. So you're already losing space there, but if you can encode particular pieces of data, of information that way, it might be actually valuable.

22:29Izar Tarandach I would have to have a good use case for it because again, it seems to me like you're putting simple bytes into multi-bytes, you're already losing space.

22:40Chris Romeo Yeah.

22:41Izar Tarandach And it's probably not going to compress nicely. So why go up when we want to go down? I mean, There must be some use for this thing. Somebody woke up one bright morning and said, hey, I'm think— I, I think I'm gonna create this, this private use areas or private use access areas in Unicode so that people can do stuff. I don't know what the stuff was that they were thinking about, but there must be some because Unicode is a standard and hopefully people looked at this and said, yeah, we agree. That's a good idea. But right now, I don't know, I'm looking at this thing and I'm saying we should have the tools to deal with this stuff. It's not, it's novel, probably, but is it something to lose sleep at 3 in the morning? Don't think so.

23:39Matt Coles You would want to check, you want to check with your, with your code linters and your SaaS tools to see if they're finding. Finding these and then update them if they're not.

23:47Izar Tarandach Yeah, definitely.

23:48Chris Romeo You're gonna, you're gonna love the reason why these things exist.

23:52Izar Tarandach Why?

23:55Chris Romeo Uh, corporate icon fonts. Companies frequently store custom icons in PUA code points, private use areas.

24:05Matt Coles Apple says this is not in a, these aren't in a private use area code block.

24:10Chris Romeo I thought that said it was.

24:12Matt Coles Yeah, it says code.

24:12Izar Tarandach Yeah, no, they are.

24:13Matt Coles Well, no, it says the code prints in question are actually not in any of the Unicode private use areas despite what the article suggests.

24:24Izar Tarandach Okay, this is a very confusing article.

24:26Matt Coles So this is the, this is the Ars Technica article talking about this. So they interpreted it as non-private use.

24:37Chris Romeo Wait, maybe the, maybe the article, uh, has conflicting viewpoints. 'Cause I'm reading the invisible code is rendered with private use areas, sometimes called private use access, which are ranges in Unicode for special characters. So I think it, I don't know, sounds like it might be, but the point is like the hidden, they, they created this private use area for nothing of really any value for having a—

25:05Matt Coles oh, sorry, that was like the article itself that It was in a comment from the article. My bad.

25:10Chris Romeo So co— so corporate icon fonts, Apple logo and vendor symbols, historic or unencoded scripts, corporate internal documents. Wait, this is getting weird. Specialized typography, emoji before standardization. They used to keep emojis in the PUA.

25:27Izar Tarandach Mm-hmm.

25:28Chris Romeo Music, mathematics, and notation systems. So like it basically exists for no real reason. Of any legitimate—

25:36Matt Coles No, it exists for a legitimate reason. The legitimate reason is we didn't think of everything, and so we need a place to put other things that we didn't think of yet.

25:44Izar Tarandach Well, apparently now it was given a legitimate reason.

25:47Matt Coles Yes, it's hacking code now.

25:51Izar Tarandach Yeah, it's too high school.

25:52Chris Romeo This is why we can't have nice things.

25:55Matt Coles Exactly.

25:56Chris Romeo Yes, it's where it always comes back to.

25:58Matt Coles It always comes back to variations or inconsistencies in a spec. Anytime you see undefined behavior or put your stuff here, bad things happen.

26:10Izar Tarandach I think that the expression you're looking for is unintended consequences.

26:12Matt Coles Yes, yes. Things like divide by zero, or there's a host of those in C, like undefined behavior between compilers.

26:23Chris Romeo Okay, kids listening at home, when you think you have a bright idea about something that's going to revolutionize the future, don't do it.

26:30Izar Tarandach You don't.

26:31Chris Romeo Okay. Because probably the answer is bad idea.

26:36Izar Tarandach Call us. And very much like somebody that we really, really like and respect, all you're going to hear is, uh-huh, but, uh-huh, but, and believe me, nothing like a but to drop your idea in place. Oh yeah.

26:53Matt Coles That's worked. That's worked really well in the past.

26:56Izar Tarandach Right? Yes.

26:57Matt Coles Yes. But.

27:00Izar Tarandach I love it that we have some proponents of yes and, and some very strong proponents of yes, but, and we pendulum between both of them.

27:10Chris Romeo You forgot about the most important response. That's the stupidest thing I've ever heard in my life. No, no, no, no.

27:16Izar Tarandach That's what everybody's thinking. I'm talking about what everybody's saying. Just say it.

27:21Chris Romeo Just say it. Be transparent.

27:22Matt Coles I think there's an opportunity for you to learn and grow with this.

27:26Chris Romeo Dude, have you ever tried that on anybody?

27:28Matt Coles Yes.

27:29Chris Romeo It worked.

27:31Izar Tarandach Wait, wait, wait, what did you try?

27:33Matt Coles This is an opportunity for you to learn and grow.

27:36Chris Romeo I feel like somebody would give you the finger if you said that.

27:39Izar Tarandach I feel like he tried that on me.

27:41Chris Romeo That was actually a mind trick on Izar.

27:45Izar Tarandach And it worked.

27:47Matt Coles You were an early victim, yes.

27:48Chris Romeo Yeah.

27:49Izar Tarandach Oh gosh.

27:52Chris Romeo Wait, did you used to work for Matt?

27:55Matt Coles I was. Yes, he did.

27:56Chris Romeo You worked for Matt?

27:59Izar Tarandach Yeah.

27:59Chris Romeo Ah, okay. All right. We'll have to have a special episode without Izar so we could talk about his work performance.

28:08Izar Tarandach Wait, let's, let's just put it like this. The word surprising, the word surprising was used in the review.

28:16Chris Romeo Wow. I've never seen that happen.

28:18Matt Coles Challenging employee.

28:19Chris Romeo I can believe that.

28:21Izar Tarandach No, I'm not.

28:24Chris Romeo Argue about that. All right, well, I think we landed somewhere.

28:31Matt Coles Yeah, it's just don't do it. It's the inverse of that other, of that shoe brand. Don't do it. We need a t-shirt. We need a t-shirt with a good logo.

28:42Izar Tarandach Just don't do it.

28:43Matt Coles Just don't do it.

28:44Izar Tarandach Oh wait, let me write that down.

28:48Matt Coles That's clever.

28:48Izar Tarandach That's actually It goes together with my, uh, just because you can't doesn't mean your AI should.

28:55Chris Romeo Yeah.

28:56Matt Coles Yeah.

28:56Chris Romeo That's true.

28:57Matt Coles Well, apparently that, that, that goes for Unicode, invisible Unicode characters.

29:03Chris Romeo Yeah. With somebody will get that on a shirt.

29:05Matt Coles That's too, if you put it, if you put a string, if you put a constant, you know, string with quotes with, with missing characters.

29:12Chris Romeo No, no, just put, no, here you go. Put the, put the invisible Unicode characters in the string. They'll never see him coming.

29:18Matt Coles They'll never see it coming.

29:20Chris Romeo Yeah.

29:20Izar Tarandach I'm sorry, but what I'm thinking about—

29:22Matt Coles What you don't know about can hurt you.

29:24Izar Tarandach I'm sorry, but I'm thinking about the logo of a very, very well-known agent maker.

29:31Matt Coles You don't need to take notes, César. It's now recorded for posterity here in this podcast.

29:35Izar Tarandach Yeah, but if—

29:36Chris Romeo Oh wait, we were supposed to hit record.

29:39Izar Tarandach I'm old.

29:40Matt Coles It's been a while.

29:41Izar Tarandach I know, but I'm still thinking about my social media manager is DevNo. Is that a good one or not?

29:52Chris Romeo It doesn't really make sense.

29:54Izar Tarandach I think so.

29:55Chris Romeo Because you wouldn't have any posts.

29:58Izar Tarandach Yeah. Wouldn't that explain a lot?

30:03Matt Coles What social media are you using that you can actually tell other people?

30:07Izar Tarandach DevRandom. My social media manager is DevRandom.

30:11Chris Romeo DevRandom is crap. There's no— it's terrible. Okay, you're random. That's pretty crappy too though, isn't it?

30:18Matt Coles I seed my RNG with invisible characters.

30:22Chris Romeo Oh, maybe that's why they exist. Oh, seeding an RNG.

30:27Matt Coles And that's an interesting, that's an interesting attack. That's an interesting attack. You have rand prends with nothing inside. As a code review, how would you, you wouldn't know it would be a fixed string.

30:36Chris Romeo You'd have, yeah, but it would cause you to freak out.

30:38Izar Tarandach I'm sorry. You just missed a really, really good t-shirt here. I see invisible characters.

30:45Chris Romeo With a picture of Haley Joel Osment on it.

30:50Matt Coles Wait, wait. Okay. You can, you can, you can modify that to just go with, I see dead code.

30:55Izar Tarandach Wait, don't we have that one?

30:58Matt Coles Do we?

31:00Izar Tarandach 0xdeadcode? I think I saw that one once, probably on Slashdot.

31:06Matt Coles Maybe.

31:07Izar Tarandach Ooh, remember Slashdot when it was good?

31:09Chris Romeo Yeah, I just, I hadn't thought of that in a long time.

31:12Izar Tarandach Remember?

31:13Matt Coles Remember? Remember?

31:16Izar Tarandach I remember.

31:17Matt Coles Until it was replaced with a little shell script.

31:20Izar Tarandach Yeah. Okay. So the next one in line is—

31:27Chris Romeo I don't know. Maybe we should, we should probably save that for another day.

31:30Matt Coles Yeah.

31:30Chris Romeo You think so?

31:31Matt Coles Yeah.

31:31Chris Romeo I think so too. We're already, uh, we, we're, we've already captivated and entertained our audience for a lot of time.

31:39Izar Tarandach We haven't dropped anybody from any running treadmills in a long time. Well, we need to do—

31:46Chris Romeo probably not till this episode. I mean, this is riveting television. Invisible Code and you.

31:51Matt Coles But I have to say, you know, when we do make it to that other article, we have to talk about forcing backgrounds to be the wrong color. When you want a dark mode, you don't need to get blasted in your face with Bah, that's my browser doing that. Not like the lights in my room.

32:11Izar Tarandach Wait, you don't have the dark mode extension?

32:15Matt Coles I have dark theme running.

32:17Izar Tarandach And no, there's an extension for Chrome-based browsers that forces too white stuff, too bright stuff into dark mode. That's actually good because I can see it in our glasses.

32:33Matt Coles Right, right. And so, you want to turn on the lights?

32:38Chris Romeo This completes our new segment called Extension Corner, where provides a recommendation on a Chrome-based extension that you think that you really need to look at. Extension Buddy, I like that right there.

32:55Izar Tarandach That's the only one No, it's the name of an extension that you should look into. We'll save that for next week.

33:01Chris Romeo There's a bonus. Now he doubled up extension corner. We don't have anything else to talk about next week.

33:05Izar Tarandach The first two ones are free.

33:08Matt Coles We had, we had a third, but we couldn't see it.

33:10Chris Romeo That's the drug dealer method of extension buddy or extension corner is where you, the hidden extension, you get away with free and see who comes back for more.

33:18Matt Coles This is the invisible extension.

33:19Izar Tarandach Yeah.

33:21Chris Romeo All right folks, welcome or fuck. I've been away from this for a long time. Welcome to, and now we're actually gonna start.

33:31Izar Tarandach No, no, no.

33:31Chris Romeo Welcome to the end.

33:32Matt Coles Oh, now you're gonna hit record.

33:34Chris Romeo Great. Yeah, we're, uh, I promise next episode we will be just as irreverent and unprofessional as we were in this.

33:42Izar Tarandach But wait, wait, wait. There's more. There's more. We approaching the 100. We, we need ideas from our, uh, uh, from our folks, our audience, from our subscribers.

33:52Chris Romeo Australia, maybe they can get together for a watch party.

33:54Izar Tarandach Yeah, from the folks in Australia. What, what they want to see on the 100.

33:57Chris Romeo Watch party, Australia is not that big of a continent.

33:59Izar Tarandach Oh, and that reminds me, there's only one Security Table. The other one is not a Security Table. This one is the Security Table.

34:06Chris Romeo This is the original. I should have started with that. Let's start over. Welcome to the original Security Table. We are the original. We've been, I had this idea 5, 6 years ago for the, for the Security Table because somebody else started a podcast called The Table. And I was like, oh, that's kind of a cool idea. Just a few people sitting around having a conversation in this, this kind of—

34:27Izar Tarandach so you see that, that, that's how things happen. Somebody has an idea. It's a good idea. You build on that idea. You don't go, oh, those guys have a security table. That's a great idea. I'm going to start a security table. That's not how it works.

34:39Chris Romeo So we're officially feuding with the other morning show. Let's send a bunch of pizzas to their house. Arrgh, me mateys. Do all kinds of wacky morning show things.

34:55Izar Tarandach Let's send them some invisible characters. There you go. Put them in an email.

35:00Chris Romeo They'll never see them. Never see 'em coming.

35:02Matt Coles White, white on white text.

35:04Chris Romeo We'll show them.

35:05Izar Tarandach And right now they just got more promotion than they ever have.

35:09Chris Romeo So that's true.

35:10Izar Tarandach You guys can thank us later.

35:11Chris Romeo Yeah. Bills in the mail. All right. Have a great rest of your day, your week, your month, wherever you find yourself on planet Earth. Keep things secure and keep rocking in the— wait, wait.

35:22Matt Coles What happens if you're on the moon or in space? You, wait, the people in the International Space Station are not allowed to have a good day.

35:31Chris Romeo Well, I hope they're not allowed to have Unicode, invisible Unicode characters. No, but, but, 'cause they're in a controlled system.

35:36Matt Coles If anyone needs them, it's probably them.

35:38Izar Tarandach But you know that in there it's like, good morning, good morning, good morning, good morning, good morning. Because as it moves, it just keeps going. It's a new day.

35:46Chris Romeo It's a new day.

35:48Matt Coles If anyone on the International Space Station is listening to the, or watching this podcast, please leave a comment.

35:54Chris Romeo Send us a message, please. That would be so cool.

35:57Matt Coles Is anybody out there?

35:58Izar Tarandach No, don't. You probably have more important stuff be doing.

36:01Chris Romeo That's true. Pay attention to the rotating, I don't know, something that controls the thrust or something.

36:12Izar Tarandach All right.

36:12Matt Coles With that, I can see the security table from my house.

36:17Chris Romeo With that, we say adieu. Adieu. Of course we do.

36:23Matt Coles Adieu.

Transcript supplied by AssemblyAI. It may contain transcription errors.

Originally published as The Security Table. Part of the AI Security Table archive.