YouTube2Text

Software Supply Chain Security: What AI Coding Changes — Transcript

by IBM Technology · 1,540 words · 95 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Let's say that you're building an awesome new app.
  2. 0:02You've built out the features, the tests are passing and you're ready to ship. Woohoo!
  3. 0:08But how much of what's about to run in production did you actually write?
  4. 0:12Well, for most apps, it's actually not a lot. Most software today isn't written from scratch,
  5. 0:17but put together from a whole bunch of pieces that came from somewhere else.
  6. 0:22So, let's fill in the rest of the picture, because you're likely to have open-source software packages,
  7. 0:28and each of those packages are pulling in their own dependencies.
  8. 0:32You've also got a container base image. This allows you to run that software in a variety of environments. My computer. Your computer.
  9. 0:41Somewhere in the cloud. You've also got a build pipeline. This is your continuous integration and deployment
  10. 0:49that allows you to constantly make updates and changes in real time.
  11. 0:52You've also got, for example, secrets that might be used to connect your application to data sources, and much more.
  12. 1:01And more and more, you also have the idea and concept of AI-generated code that a model is creating and writing.
  13. 1:09So every one of these is something that your software is depending on.
  14. 1:14And most of them you didn't write and some of them you probably never even looked at. I'm guilty of that one too.
  15. 1:21But this whole picture is what's known as the software supply chain.
  16. 1:34It includes the application itself, but on the other hand, everything else around it, as well.
  17. 1:45In fact, you've probably heard of popular examples of supply chain security attacks, for example, with software bugs like the incident of log4j.
  18. 1:56Now, Log4j is a popular logging library that was in almost every piece of software in order to collect metrics and logs,
  19. 2:03and almost 93% of enterprise environments had it installed as a dependency in the software they use, or maybe we're using as a service.
  20. 2:12And a security flaw left teams vulnerable to remote code execution.
  21. 2:17That means if I have my server here and a malicious attacker was to use a specific command, they could get access to our environment to exfiltrate data,
  22. 2:29to sabotage things, and remediating this issue was quite difficult because it was in not just servers, but applications that you and I use every day.
  23. 2:38And it was in the hybrid cloud, uh, deployed on thousands of different machines.
  24. 2:42So that's an example of software bugs where remediation of a dependency can be quite difficult.
  25. 2:47There's also issues of ecosystem poisoning.
  26. 2:50So you might have heard of attacks on different package registries like NPM for JavaScript developers or PyPI
  27. 2:58for Python developers, where attackers intentionally upload malicious packages either in typo squatting,
  28. 3:06where let's say I'm looking for a specific package called request in order to make requests from my app to another,
  29. 3:12and someone has taken a name of request, but with a small typo, maybe two S's. And if I type this in, it's in my application.
  30. 3:22I could get hacked.
  31. 3:23There's also the idea of hacking a legitimate account of a contributor to a popular project,
  32. 3:30and using their account to upload a new version of software that might be downloaded automatically on your or I's application,
  33. 3:40hich could be vulnerable. Andfinally, there's attacks through the build pipeline, ah, of software.
  34. 3:46So there's a popular example of SolarWinds, which is a distributor of software where they had attackers that breached their environment
  35. 3:56to covertly, um, hack about 18,000 of their customers with a backdoor. And even though
  36. 4:03it was a trusted software provider for these different customers,
  37. 4:06well, that backdoor got in because of a code build process and was updated automatically on all those different environments.
  38. 4:14So, these are different entry points for your software supply chain. Some are through a dependency or an ecosystem. Some are through the build process.
  39. 4:23But this is really why we have to care about the whole picture.
  40. 4:26So none of this is necessarily brand new. But what has changed?
  41. 4:30Well, AI has changed things from both sides actually, where on one side,
  42. 4:34the models and agents are now producing much more code that we're actually using in applications
  43. 4:40and senior engineers now get dozens of different review requests per day.
  44. 4:45And trust me, people were already struggling to keep up before AI.
  45. 4:48On the other side, though, AI is finding vulnerabilities and helping us test and validate our applications as we deploy them.
  46. 4:57And these frontier models can scan open-source code and dependency chains at machine speed. And that means the time between a bug being disclosed
  47. 5:06and someone using it in attack, or you having to prevent that, is getting a lot shorter.
  48. 5:11So I like to think of AI and agents as a new supplier in our supply chain.
  49. 5:16It's a really fast one, and just like any other supplier, you need a way to check what it ships.
  50. 5:22A lot of teams are still finding vulnerabilities way over here near the end, and by then fixing it can be quite disruptive because
  51. 5:29you have to reload all the context, rework something that's maybe already been merged,
  52. 5:34and hold up a new release. It's like when you're deep in a problem and you get pulled into a meeting
  53. 5:39and you come back an hour later with no idea where you were.
  54. 5:41Except now the code is weeks or months old. And so that's why you hear this term of shift left. And what does this really mean?
  55. 5:50Well, this means checks as you're building out new features before and after you push code, signing
  56. 5:57off on those artifacts in the continuous integration pipeline before you get to production
  57. 6:01And really as many checks as possible that can be automated and done up front so that the real
  58. 6:07attention from a person or an AI reviewer goes to the activities that really need judgment.
  59. 6:14So what can you start doing today
  60. 6:16Well, there are four simple things. And first is to know what's in your application.
  61. 6:21This is exactly what an SBOM is for. It's a software bill of materials,
  62. 6:26and you can think of it as the ingredient list for your application or software.
  63. 6:30That means when the next Log4j or big critical vulnerability shows up,
  64. 6:34all you have to do is check the list to see what's in your application and which version instead of going hunting in those applications that you might manage.
  65. 6:43And second is to check early and check automatically.
  66. 6:47This means to scan the dependencies that you're using for critical vulnerabilities, to scan for secrets that might be exposed,
  67. 6:55and to validate authenticity of those dependencies and code that comes in.
  68. 6:59So, speaking of code, keep those changes small. A giant change that mixes a dependency bump,
  69. 7:05a refactor of code and new behavior for your application is hard for anybody to review,
  70. 7:10whether it be human or AI.
  71. 7:12And speaking of AI, verify independently.
  72. 7:16So if you're using an AI agent for code generation or code review, this means that if you're using the agent to
  73. 7:25both write the code that is going into the application and review the code to make sure that it's functional, well, it's likely to miss the same thing twice.
  74. 7:35So, a good review should use a separate model or a separate session so that you have this kind of separation of concern and zero trust.
  75. 7:44And this is a proven idea that we already use in a lot of other fashions.
  76. 7:49But, finally is to prove where your application came from. So this means to sign your artifacts and
  77. 7:57code so that you can show provenance.
  78. 8:00And there's actually already a framework for this called SLSA
  79. 8:03from the Linux Foundation, which is really good to follow here because it gives you a way to record
  80. 8:10what was checked in your application and document why it was approved.
  81. 8:14As your application grows larger and larger, this gets more and more important. So even if you do all four of these,
  82. 8:21something will get through eventually, and maybe you'll fix it by patching or upgrading the software in your application from,
  83. 8:29say for example, version one to version two.
  84. 8:33But in a lot of situations, maybe it'll break your application and you just can't do it So that's where initiatives to combat open-source vulnerabilities come in.
  85. 8:42So there's two of them I want to mention Project
  86. 8:45Akrites from the Linux Foundation and Lightwell from Red Hat and IBM,
  87. 8:50which have fixed packages of software ready to go so that you don't have to automatically upgrade,
  88. 8:56which might break your application.
  89. 8:58So the thing is, software supply chain security, it's not about slowing things down,
  90. 9:04but catching these issues before they get to production so that you're able to go ship faster
  91. 9:09and feel confident about it.
  92. 9:11You want to be able to think that, hey, if something were wrong, how would we know?
  93. 9:15And I believe that after watching this video, you can answer that question.
  94. 9:19So, thanks so much for watching. If you learned something today,
  95. 9:23be sure to like the video and subscribe to the channel. We'll see you in the next one.

About this transcript

This page contains the full transcript of Software Supply Chain Security: What AI Coding Changes by IBM Technology, generated from the public captions YouTube serves with the video. The transcript has 1,540 words across 95 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.