How Security Tools Must Evolve – Dan Kuykendall – ASW #261

Noting this briefly to highlight the how monitoring has been critical in detecting compromises in SaaS providers.

Cloudflare published a similar blog post last week about the Okta incident. They've also now provided an open source HAR sanitizer in response to the technique attackers used to obtain session tokens.

Last week we talked about OAuth and this week we have some more OAuth implementation mistakes.

These stem from not verifying tokens during the OAuth dance. Thus, attackers who were able to insert their own tokens could consequently impersonate others.

It's an example of how a secure design meets an insecure implementation. And leaves us with a thought exercise in how the design might be improved so that token validation isn't just required, but is implemented by default due to the nature of the protocol or its cryptographic properties.

CVSS v4.0 is slated for publication this week. Be ready on November 1st to remember the distinction that CVSS Base metrics represent severity, not risk, and that you can have some CVSS-BE and CVSS-BTS along with that CVSS-B.

Ok, this is the second PDF this month behind a registration wall, but it's a good gauge of where to strive for maturity in your DevOps teams.

This time around I wanted to focus on its attention to user-centrism and quality documentation. User-centric features and approaches to software reap benefits for internal as well as external tools. I'm also a fan of good, concise documentation -- but I also feel like that's an area that is under-recognized and not incentivized in engineering promotion paths.

Cyberpunk's aesthetics are cool, but the future behind it is always dystopian. Here's a cyberpunk-esque fight between artists and AI training models where artists are using techniques to subtly poison those models to prevent their work from being co-opted.

Check out the Glaze project that the team also created and will be integrating Nightshade's techniques into.

Here's an implementation of a spectre attack in JavaScript and WebAssembly. It works against WebKit (Safari) on recent Apple chips. There are many interesting angles on this, from side-channels to sandboxes to LockDown mode. It's also a reminder that some vulns are difficult to exploit, but can have high impacts and be just as difficult to counter.

Check out the lab's other work on hardware security. It's all very well written and they got a perfect domain name, architecture.fail.

There's a great article from Dan Goodin here.

I feel like we haven't been talking enough about software design and design examples. So here's one article to help shift that balance!

It talks about some coarse sandboxing, like VMs and containers, but also covers seccomp.

This is one of the longer articles we've included and it spans quite a bit more than just appsec. But it's an educational read and I like the points it makes when it talks about the framing of security and the anti-patterns of blaming users for clicking on links.

There's a lot more to the article than that, so grab some tea and sit down for a long read.

Three critical vulnerabilities were announced last week about the nginx-based ingress controller for k8s. To me, the most significant is the ability to perform arbitrary code execution through injection into a configuration-snippet annotation.

The article link is to one of the three - go up a level in the kubernetes-announce archive and you'll see the others.

The computer security group at the Swiss Federal Institute of Technology in Zurich have created a fuzzer for RISC5 CPUs, and found a 37 new bugs with 29 new CVEs. Besides being Yet Another CPU Fuzzer, what's interesting about this one is it's designed to create code that should execute successfully if there is no issue, whereas in many fuzzers they're a little more willy-nilly in their attempt to crash the target. This means for Cascade, it should be easier for researchers to determine bugs and troubleshoot failure cases

This landed on my radar from of all places a conversation on Nextdoor. In previous years, they used to do testing competitions, similar to capture-the-flag competitions. This sounds amazing, and I hope they bring it back.

In the meantime, I think almost any of our companies can benefit from working on improving our testing processes, so there's probably lots to learn here.

I don't think this is exploitable, but still I find these type of bugs "fun" when they're found in software packages that have been around for a while.

Patch is at https://github.com/vim/vim/commit/9198c1f2b1ddecde22af918541e0de2a32f0f45a - a little more complex that I would have expected...

One of the things I like machines for is doing work that humans can't - or shouldn't - do. One example of this might be removing "dead code." Are you SURE that code isn't being used anywhere? Meta's SCARF takes a number of steps like analyzing code graphs and call logs to ensure it's able to say "yes," and then sends a pull request.