Jim Simpson

SANS Certified Instructor Candidate. All views are my own, likes and retweets are not always endorsements.

[HiddenLayer @hiddenlayersec]

🔍 Can a single image hijack your AI’s behavior?
Yes & without changing the application.

Meet VISOR: a new method that steers GenAI models using images alone.

It’s a new class of AI vulnerability and a new opportunity for AI alignment.
🔗https://t.co/Mv8ENWQR72

[Jim Simpson @4a696d]

So this one has been a while coming. If you've seen any of the talks or workshops I've done recently you'll have heard me talking about the way we are thinking about attacks on generative AI. Well, it's finally here.
https://t.co/7LCM2JQQ8R

[HiddenLayer @hiddenlayersec]

🔐 Not all prompt injections are the same.

We just released a taxonomy of adversarial prompt engineering, mapping the why, how, and what behind LLM prompt attacks.

Built for red teamers, defenders & researchers. Open to the community.

🔗 https://t.co/LOmXs1sZfo

[HiddenLayer @hiddenlayersec]

HiddenLayer researchers have found a way to bypass text classification models by targeting tokenizers. TokenBreak gets past protection models, leaving end targets exposed.
🔗 https://t.co/JdWLwW2rN9

#AISecurity #AI #LLMSecurity

[mRr3b00t @UK_Daniel_Card]

Still need to work on this loads more but in real world testing this is being useful!

#CyberCrime #Mapping #Graphs #IOCs #Enrichment #OSINT #ThreatIntel #Intel #Crime #Cyber #hacking

https://t.co/Ycwzt2u5Q5

[IISF @iisf_ie]

IISF May Chapter Meeting #IISF #SANS #CyberSecurity Understanding LLMs

[HiddenLayer @hiddenlayersec]

New on the AI Engineering Podcast: HiddenLayer’s Kasimir Schulz breaks down how to secure ML models from shadow logic & model ancestry to scanning for threats.
-Scan models
-Trace neural subgraphs
-Know your attack surface
Listen here: https://t.co/BTZiysM6ee
#AIsec #MLSec #AI

[AISecHub @AISecHub]

Critical AI Security Guidelines by SANS Institute

🔐 Access: Use least privilege + zero trust. Lock down models, vector DBs, and APIs.

📦 Data: Protect training + augmentation data. No sensitive info in prompts. Sanitize everything.

🚀 Deployment: Avoid hardcoding rules into LLMs. Use RAG + ACLs. Vet public models for backdoors.

🧠 Inference: Validate inputs and outputs. Guardrails help, but can't be your only defense.

📊 Monitoring: Log prompts, watch for drift, detect abuse. Monitor API usage closely.

⚖️ GRC: Track AI usage, align with evolving regs (EU AI Act, EO 14110, etc). Build AIBOMs + model registries.

Authors: Ahmed AbuGharbia (Fortinet), Sarthak Agrawal (U.S. Congress), Brett D. Arion, CISSP, (Binary Defense), Matt Bromiley (Prophet Security), Ron F. Del Rosario (SAP), Mick Douglas (InfoSec Innovations), David Hoelzer (Occulumen), Ken Huang, CISSP (https://t.co/r9AyVkowT4), Bhavin P. Kapadia (Stablecoins), Rob T. Lee (SANS), Seth Misenar (Context Security Consulting), Helen Oakley (SAP), Jorge Orchilles (Verizon), Jason Ross (OWASP AI), Rakshith Shetty (Palo Alto Networks), Jim Simpson (HiddenLayer), Jochen Staengler (BSI), Rob van der Veer (Software Improvement Group), Jason Vest (Binary Defense), Eoin Wickens (HiddenLayer), Sounil Yu (Knostic).

@aagsec, @sarthak777, @barion, @_bromiley, @SAP, @bettersafetynet, @it_audit, @kenhuangus, @robtlee, @sethmisenar, @e2hln, @jorgeorchilles, @OWASP, @PaloAltoNtwks, @4a696d, @BSI_Bund, @robvanderveer, @Binary_Defense , @enwckns, @sounilyu, @SANSInstitute

More info: https://t.co/YspiEje1Mm

#AIsecurity #Cybersecurity #LLMs #GenAI #AIgovernance #AIethics #Infosec #SecureAI #AISafety #SANSInstitute

[HiddenLayer @hiddenlayersec]

New Research: New Gemini for Workspace Vulnerability Enabling Phishing and Content Manipulation. Our latest blog examines vulnerabilities in Google’s Gemini for Workspace, a versatile AI assistant.
🔗 Read more: https://t.co/dq9IYsAFEx

[Microsoft Threat Intelligence @MsftSecIntel]

Microsoft identified a North Korean threat actor exploiting a zero-day vulnerability in Chromium (CVE-2024-7971) to gain remote code execution. Our assessment of ongoing analysis and observed infrastructure attributes this activity to Citrine Sleet. https://t.co/ITqOQpWn2v

[Jon Gorenflo 🇺🇦🌻 @flakpaket]

I’m looking for success or horror stories from people or orgs that have tried to disable IPv6 in Windows across the enterprise. Aaaaand…GO!

[SpecterOps @SpecterOps]

Apeman is a tool designed to simplify the understanding of permissions and potential attack paths within an AWS environment for both attackers and defenders. Join @hotnops' Arsenal session at #BHUSA to see this tool in action.

Learn more 👉 https://t.co/B2ZSsrQ7dW

[Binni Shah @binitamshah]

BlueSpy - PoC to record audio from a Bluetooth device : https://t.co/sv0peS2LhC credits @Tarlogic

[Koen Van Impe ☕ @cudeso]

Kick off the week with a new @MISPProject playbook! Query Elasticsearch @elastic for threat intelligence from custom MISP searches, plot the results in a graph, report sightings in MISP, and send a summary to Mattermost. #cti #automation #playbooks https://t.co/NzaCjO7mWz

[Taz Wake @tazwake]

Most (if not all) #Linux Distros have moved to the Journal with their latest versions. This is a LinkedIn article on key points for DFIR.

If you'd like to know more, check out FOR577 from SANS (https://t.co/UTqxJjd59M) - attend in-person or live online!

https://t.co/PJGUDLHlal

[Kathryn Hedley @4enzikat0r]

If you enjoyed my first #GettingStartedInDFIR workshop last week, make sure you’re also signed up for the second workshop on July 2 where we’ll explore file extensions, signatures & different types of file metadata

#DFIR @sansforensics

[Kathryn Hedley @4enzikat0r]

If you haven’t yet signed up to the first of my #GettingStartedinDFIR workshops, do it now!

Friday 21 June 13:00 EDT / 17:00 UTC
Join live or get the recording to watch & play along later

[Pratyush Maini @pratyushmaini]

1/We've nailed a framework to reliably detect if an LLM was trained on your dataset: LLM Dataset Inference.

After over a year of thinking of writing about how hard this is, we had a breakthrough that made me quite literally jump from my seat!

📝: https://t.co/B9jvMsoNGW Long🧵

[Jim Simpson @4a696d]

@attrc @josephfcox Yep, this!