Becoming the trainer: Attacking ML training infrastructure Learn more about machine learning training environments and infrastructure, as well as different attack scenarios against critical components, including cloud compute, model artifact storage and model...

New research just dropped I'll be presenting at @wearetroopers.bsky.social next week - Attacking ML Training Infrastructure
💥 Model poisoning for code execution
⚠️ Abusing ML workflows
⚙️ MLOKit updates and new threat hunting rules

www.ibm.com/think/x-forc...

Black Hat Black Hat

Register while you still can for @retbandit.bsky.social and I's @blackhatevents.bsky.social #BHUSA training, seats are filling up fast!!
www.blackhat.com/us-25/traini...

I am thrilled to be presenting new research on attacking ML training infrastructure at @wearetroopers.bsky.social this summer. Stay tuned for a blog post and lots of updates to MLOKit closer to the conference!

Black Hat Black Hat

Learn 📝 about this emerging topic in a first-of-its-kind #BHUSA training from @retbandit.bsky.social and I where you will use hands-on labs to perform attacks such as model theft, model poisoning and much more 🤖

blackhat.com/us-25/traini...

Fileless lateral movement with trapped COM objects | IBM New research from IBM X-Force Red has led to the development of a proof-of-concept fileless lateral movement technique by abusing trapped Component Object Model (COM) objects. Get the details.

[Blog] This ended up being a great applied research project with my co-worker Dylan Tran on weaponizing a technique for fileless DCOM lateral movement based on the original work of James Forshaw. Defensive recommendations provided.

- Blog: ibm.com/think/news/f...
- PoC: github.com/xforcered/Fo...

I am excited to announce the first conference dedicated to the offensive use of AI in security! Request an invite at offensiveaicon.com.
Co-organized by RemoteThreat, Dreadnode, & DEVSEC.

The Security Conversation - The value of offensive security work is fully realized by participation in the security conversation.

aff-wg.org/2025/03/13/t...

#MythicTip Want to start automating stuff with Mythic, but not sure where to start? Check out the built-in Jupyter notebooks with Mythic Scripting installed and have fun! Lots of ready to run examples exist already :) Just log in with the Jupyter token from your .env file

It was an honor to speak at the @780thmibdecyber.bsky.social’s AvengerCon on the use of AI in Offensive Cyber Operations, Vuln Discovery/Weaponization, OST Dev as well as attacking AI systems.

Here’s a few slides from the talk…

@NSACyber @ARCYBER @CISAgov @US_CYBERCOM

Normally you can't auth to Entra ID connected webapps with bearer tokens. But if Teams can open SharePoint/OneDrive with an access token, I guess so can we. roadtx now supports opening SharePoint with access tokens in the embedded browser 😀

New blog post on the abuse of the IDispatch COM interface to get unexpected objects loaded into a process. Demoed by using this to get arbitrary code execution in a PPL process. googleprojectzero.blogspot.com/2025/01/wind...

Black Hat USA 2025 Black Hat USA 2025

If you would like to learn how to attack and defend popular platforms that are used to develop and deploy ML models, early sign-up is now available for @retbandit.bsky.social and I's @blackhatevents.bsky.social training course ⬇️

www.blackhat.com/us-25/traini...

Intune Attack Paths — Part 1 Intune is an attractive system for adversaries to target…

In Part 1 of my Intune Attack Paths series, I discuss the fundamental components and mechanics of Intune that lead to the emergence of attack paths: posts.specterops.io/intune-attac...

ShmooCon 2025 - YouTube You can reach me at https://twitter.com/Strong1Wind

Live streams from the last ShmooCon security conference, which took place last week, are available on YouTube

www.youtube.com/playlist?lis...

Thank you very much, glad you enjoyed it!

You can find our @shmoocon.bsky.social presentation slides at the below GitHub repo. Thanks again to all that attended. Also, thank you to the conference organizers for putting on a great con and having us! #shmoocon

github.com/h4wkst3r/Con...

Thank you Casey, we really enjoyed your talk as well on the Windows projected file system!!

Hijacking Azure Machine Learning Notebooks (via Storage Accounts) Abusing Storage Account Permissions to attack Azure Machine Learning notebooks

New
@netspi.bsky.social
blog out today on "Hijacking Azure Machine Learning Notebooks (via Storage Accounts)". This is very similar to Storage Account attacks that have been done against Function/Logic Apps and Cloud Shell - www.netspi.com/blog/technic...

What is Tier Zero — Part 1 Tier Zero is a crucial group of assets in Active Directory (AD) and Azure. Its purpose is to protect the most critical components by…

Unequivocally one of the best pieces of writing on Tier 0 there is...

GitHub - xforcered/MLOKit: MLOps Attack Toolkit MLOps Attack Toolkit. Contribute to xforcered/MLOKit development by creating an account on GitHub.

Whitepaper and MLOKit tool referenced in blog post:

Whitepaper 📝

www.ibm.com/downloads/do...

MLOKit Tool 🔨

github.com/xforcered/ML...

Abusing MLOps platforms to compromise ML models and enterprise data lakes With the rush to implement AI across organizations came the increase in the use of MLOps platforms and a greater risk of attack. Learn more about MLOps platforms and how threat actors are using them.

MLOps platforms are becoming critical to enterprises. This has caused @retbandit.bsky.social and I to research these platforms and how they can be abused by attackers. Check out our research we will be presenting @shmoocon.bsky.social this week.

securityintelligence.com/x-force/abus...

GitHub - logangoins/Krueger: Proof of Concept (PoC) .NET tool for remotely killing EDR with WDAC Proof of Concept (PoC) .NET tool for remotely killing EDR with WDAC - logangoins/Krueger

IBM X-Force's Logan Goins has released Krueger, a .NET tool for remotely killing EDR using the Windows Defender Application Control (WDAC) utility

github.com/logangoins/K...

Detection engineering rabbit holes — parsing ASN.1 packets in KQL TL;DR: Detection engineering is sometimes hard. Your efforts may seem to have failed, but perseverance can pay off. Or you can still fail…

Detection Engineering is sometimes hard, and may fail. Still a lot of things can be learned by the process. In this blog I cover a lot. I had a detection, currently it's broken but MS is on it :D

medium.com/falconforce/...

Black Hat Europe 2024 Black Hat Europe 2024

Secureworks Japan has released PyTune, a post-exploitation tool for enrolling fake devices into Microsoft Intune

www.blackhat.com/eu-24/briefi...

github.com/secureworks/...

@retbandit.bsky.social and I are thrilled to be speaking @shmoocon.bsky.social in January on research we have been conducting on attacking and defending popular enterprise Machine Learning Operations (MLOps) platforms we see during adversary simulation engagements. Whitepaper and tool coming soon!