📺️ Netflix’s Scott Behrens on Building a Paved Road

In case you missed it live, my chat with Scott Behrens is now on YouTube!

Scott’s been at Netflix for 11 years and has seen the security team grow from a handful of people to well over 100, and is the the strategic tech lead for all of Security, Privacy, and Risk at Netflix. We discussed:

1. Gotchas if you’re trying to build a Paved Road at your company.

2. Validating that your Paved Road / security control guarantees actually do what you expect them to.

3. Lots of audience questions!

I’d love to hear what you think, and any other questions you have. Let me know!

👉️ Watch it here 👈️ 

🦃 P.S. No newsletter next week, enjoy Thanksgiving if you celebrate.

I’ll bust back into your inbox the first week of December like the Kool-Laid Man- oh yeah!

Sponsor

📣 AI Devs Sound Off on Container Challenges

The rise of AI has led to more containerized environments, but what do AI/ML developers really think about using containers? Our latest white paper dives into insights directly from developers working with AI and ML, highlighting key challenges like bloated container sizes, "dependency hell" with version conflicts, and gaps in container knowledge that create confusion and setbacks. Discover the real struggles and possible solutions to help optimize container use in AI.

👉 Read the Report 👈

AppSec

intrudir/burpcollaborator-docker
By Jonathan Conesa: A set of scripts to install a Burp Collaborator Server in a Docker environment, using a LetsEncrypt wildcard certificate.

Paved Roads? Secure-by-Design?? More Buzzwords???
Srajan Gupta argues for the importance of "paved paths" in security engineering, which involves creating standardized, easy-to-adopt security controls and development patterns. He suggests starting small by identifying common patterns (like authentication and logging), building reusable components, and making them attractive to developers through excellent documentation and automated validations. Treat secure development as a product, with developers as customers. Start small, measure adoption, and iterate based on feedback.

“When done right, security becomes invisible infrastructure.”

“Would you rather build roads one car at a time, or create highways that enable thousands to move faster, safer, and more efficiently?”

Sponsor

📣 Overcome Key PCI DSS 4.0 Control Hurdles

Meeting critical PCI DSS 4.0 security requirements while maintaining operational agility might feel like a tightrope walk – but it doesn’t need to.

Explore our guide on how to navigate the security demands of PCI DSS 4.0 compliance without hindering engineering and development – and find out how to address key control requirements with the Teleport Access Platform.

👉 Download Guide 👈

I was actually at a CISO event recently (shout-out to Lightspeed 🙌 ) and the CISO next to me at dinner said they bought Teleport and liked it, and that it saved them some headaches.

Apple

Reverse Engineering iOS 18 Inactivity Reboot
Jiska dives into iOS 18's new inactivity reboot feature, which causes an iPhone to reboot after 3 days of inactivity to protect against forensic attacks. Jiska walks through reverse engineering the SEPKeyStore kernel extension and Secure Enclave Processor firmware to determine how this feature works. The inactivity reboot significantly raises the bar for both law enforcement and thieves attempting to access locked iPhones. Very cool work!

Pishi: Coverage guided macOS KEXT fuzzing
Meysam presents Pishi, a static binary rewriting tool designed to instrument basic blocks of the XNU kernel and macOS KEXTs. For the XNU kernel, Pishi allows you to instrument at a function, file, or folder level. The post covers implementation challenges, optimizations, and using libprotobuf-mutator for stateful APIs.

💡 Wow, an incredible amount of detail and excellent related work references 🔥 One of the most detailed fuzzing posts I’ve read, outstanding.

A New Era of macOS Sandbox Escapes: Diving into an Overlooked Attack Surface and Uncovering 10+ New Vulnerabilities
The blog post for Mickey Jin’s POC2024 presentation. Mickey describes a number of sandbox escape vulnerabilities in macOS discovered by exploiting an overlooked attack surface: XPC services in the PID domain, which are not expected to be invoked from a sandboxed application, so there are no additional entitlement checks or sandbox checks for the incoming XPC clients. The post provides exploit code and demonstrations for each vulnerability, along with Apple's patches and subsequent bypasses in some cases.

Cloud Security

xen0l/aws-gate
By Adam Števko: A better AWS SSM Session manager CLI client. Connect to instances by other means (e.g. DNS, IP, tag, instance name, autoscaling group), and it supports a configuration file where you can store connection info.

Protecting Data and Preventing Ransomware: The IAM Guide to Managing and Updating Encryption for AWS Resources
There have been a few posts recently about how attackers can conduct ransomware attacks by encrypting or re-encrypting cloud data resources with an encryption key they control. Fog Security’s Jason Kao shares research on the IAM permissions necessary and how to update encryption settings across 65 resources over 45 AWS services. One noteworthy finding is that many methods that can update encryption settings do not mention encryption. Tip: consider blocking unneeded encryption permissions with SCPs.

I bought us-east-1.com: A Look at Security, DNS Traffic, and Protecting AWS Users
Clever post by Gabriel Koo: Gabriel bought the domain us-east-1.com, a popular AWS region, and shares the DNS traffic he received, revealing frequent queries for prod-backend-db, loopback-streaming, and various AWS service-related subdomains. He also received emails from likely AWS test environments.

Recommendations: make sure you’re not using domain placeholders that could actually be routing to an attacker-controlled domain, check for typos, and consider using tools like Route 53 Resolver DNS Firewall.

Container Security

vmware-tanzu/pinniped
The easy, secure way to log in to your Kubernetes clusters. Pinniped provides identity services to Kubernetes: easily plug in external identity providers, give users a consistent, unified login experiences across clusters (on-prem and managed cloud environments), integrate with an enterprise IDP, and more.

Making Sense of Kubernetes Initial Access Vectors Part 1 - Control Plane
Wiz’s Shay Berkovich presents a taxonomy of initial access vectors for Kubernetes clusters, focusing on the control plane, covering Kubernetes API access (unauthenticated access, Kubeconfig, Kubectl proxy), Kubelet API access, and management interfaces (e.g. K8s Dashboard, Kubeflow, Argo Workflows). Shay discusses the associated risks and suggest protection and detection techniques.

Kubernetes Initial Access Vectors Part 2: Data Plane
Wiz’s Shay Berkovich discusses Kubernetes data plane access, covering potential initial access vectors originating from the applications running on the cluster, the NodePort service, concerns around container images (their origin and potential escape scenarios), and wraps up with execution-as-a-service workload types.

Supply Chain

👋 My friend Kyle Kelly, author of the great CramHacks newsletter, is looking for roles focused in supply chain security. He’s hardworking and hungry to do great things, reach out directly if you’re interested.

Repo swatting attack deletes GitHub and GitLab accounts
Recently there were a number of posts/media attention on how malicious files could be uploaded to GitHub via creating a GitHub Issue or writing a comment, for example. Paul McCarty proposes “repo swatting” as an attack in which you purposefully upload malware to a target victim repo, and then report that repo to the platform’s abuse team, who may then delete the user’s account.

Note that this attack no longer works, as GitHub changed their generated file upload URLs to be UUIDs/paths that do not contain the user or repo names.

Attestations: A new generation of signatures on PyPI
Trail of Bits announces PyPI's new index-hosted digital attestations, which is enabled by default for packages published to PyPI using Trusted Publishing, automatically providing build provenance. The post has some nice technical details and an overview of implementation choices, but the TL:DR is that it’s now relatively straightforward to have a set of cryptographically verifiable claims about a repository and workflow state corresponding to the time at which a package was published to PyPI.

Release-Drafter To google/accompanist Compromise: VRP Writeup
Adnan “GitHub Action Slayer” Khan discovered a vulnerability in the popular Release Drafter GitHub Action that could allow attackers to obtain a GITHUB_TOKEN with write access to the action's repository using the Dependabot actor confusion technique (included in a prior tl;dr sec). Adnan demonstrated how this could lead to a supply chain attack on downstream users like Google/Accompanist, who used the action by tag instead of SHA.

AI + Security

Prompt Injecting Your Way To Shell: OpenAI's Containerized ChatGPT Environment
0din’s Marco Figueroa walks through exploring OpenAI's Debian-based containerized ChatGPT environment, showing how you can list files on the file system, upload files like Python scripts and execute them, and more. He also walks through extracting custom GPT instructions (their system prompt or any base instructions) and uploaded knowledge data they use.

See also Daniel Wood’s post also examining ChatGPT’s execution environment which he published on Dec 7, 2023, which also includes examining the running processes, port scanning the internal network, etc.

Predictable IDs & PII Leakages: Using AI to Mass leak data
HackerOne’s Shlomie Liberow describes how he leveraged AI to uncover vulnerabilities in a document ID system that allowed accessing other user’s documents via custom broken document ID generation. Shlomie used Claude to analyze 100 generated IDs, identified a timestamp-based pattern, which they then cracked using hashcat. He was able to use a sandwich attack: upload a doc and get its ID, wait, request the ID again, and then use the timestamp and sequence number to work out the seed and download other user’s documents.

💡 There a couple of neat things about this post: first, giving an LLM a bunch of IDs or other data and asking for it to identify patterns/what might be generating them/potential implementations is great. Also, taking your hypotheses and having the LLM generate quick scripts that implement them so you can rapidly (in)validate what might be happening is an awesome productivity hack. Lots of great tips in this post 👍️ 

How XBOW found a Scoold authentication bypass
Nico Waisman and Brendan Dolan-Gavitt describe how XBOW, an AI-powered security testing tool, autonomously discovered and exploited a critical vulnerability in Scoold, an open-source Q&A platform.

The vulnerability allowed attackers to bypass authentication, read application configs (including API secrets), and potentially take over Scoold instances. XBOW leveraged a combination of Java bytecode analysis, fuzzing endpoints using ffuf, and creative use of the HOCON configuration language to achieve arbitrary file read on the target system.

💡 The multi-step reasoning and testing of a number of potential attack hypotheses, and pivoting when things aren’t working, is pretty cool. And figuring out how to use a custom config language to achieve a file read goal is impressive.

Misc

• Alex Hormozi on the power of prioritization, focus, and urgency. Also, his Old Spice parody commercial for Skool is 😂 

• Eric Schmidt on Diary of a CEO - Eric is super sharp, and shares interesting thoughts on the history of tech, early Google, and the future of AI

• AI music and copyright seems like a mess - there’s currently not much legal precedent, so you can use AI to generate something that sounds like an artist, and then put that into YouTube’s content ID platform, and then the artist’s future work might get marked as IP infringement 😆 

• Push-up technique tip

• Peloton instructors reading inappropriate names 😂 

• Impacts of AI on Hiring, What to Do if You Get Laid Off, & Being a Dad | Prof Galloway Office Hours - I like the framing of part of becoming an adult/man is providing surplus value to your family, community, and the world, and looking out for people.

• 🧵on the difficulties of legal immigration in the U.S. - Personally, I think we should (for the most part) welcome any highly skilled person who wants to live here.

• Lenny’s Podcast: Building Wiz: the fastest-growing startup in history | Raaz Herzberg (CMO and VP Product Strategy)

• How Wiz Became the Fastest Software Company to Hit $500M & Its Path to $1B by Francis Odum

✉️ Wrapping Up

Have questions, comments, or feedback? Just reply directly, I’d love to hear from you.

If you find this newsletter useful and know other people who would too, I'd really appreciate if you'd forward it to them 🙏

Thanks for reading!

Cheers,
Clint
@clintgibler