Pinned Tweet
After coining the term MLSecOps in 2017, I'm finally presenting the best you ever saw introduction to MLSecOps, or DevSecOps for AI systems, with core principles, ML pipeline stages, and examples!
Slides and Video: conf42.com/DevSecOps_2022_Eu…
#AI #SecureAI #MLSecOps #DevSecOps
Introducing the AI Risk Quadrant with 100 agent security profiles!
📈 AI Risk Quadrant: 100 agents across 10 agent classes
📊 AIRQ Framework: attack surface vs. defense controls
🗂️ AIRQ Agent Profiles: security analysis of 100 agents
Read more: AIRiskQuadrant.com
1
2
140
The irony of the AI red teaming market is that vendors who aren't completely sure what product to build are selling it to customers who aren't completely sure what product they need.
3
217
The immediately jailbroken "new advanced AI protections" announced by both Anthropic and Google are a perfect example of the Security Theater concept.
2
229
I got a 64gb RAM MacBook Max to run local LLMs, but instead I'm running Chrome with 791 tabs -- a lot of AI Security and product strategy stuff.
I guess just sharing all those links is like 2 years' worth of content!
2
170
Using AI responsibly is like riding a bicycle
The only meaningful way to make it useful is:
-> Take full ownership of riding the bicycle to the destination
-> Expect zero help from the bicycle beyond your own effort
-> Assume no responsibility for the outcome from the bicycle
1
110
The AI security market is like the early 2000s in terms of attack-defense coverage.
It should catch up and converge with the broader security market, though.
If every software will become AI software, is there any difference?
1
118
I miss the times when you only needed a single weekend to catch up on an entire year of progress in AI security and waited weeks or even months for any big news!
4
181
What are technically cool but practically unsolvable security problems?
My pick: the classic Adversarial ML.
1
6
250
I asked ChatGPT to roast me and ended up with the best resume for my AI Security consulting practice!
1
213
Can't wait to see how we transition from trustworthy AI to zero-trust AI -- just like the shift from trustworthy computing to zero-trust security.
P.S. Here's a 22-year-old column from Bill Gates introducing Trustworthy Computing:
wired.com/2002/01/bill-gates…
Very often it's unlikely I'll find good content when a title includes both "Security for AI" and "AI for Security."
1
4
313
Cast your vote and help OWASP choose the next Top 10 risks for LLM security!
The voting process will also eliminate some entries, keeping only the top 10 -- so choose wisely.
Only a few days left!
Contribute your opinion via this Google Form:
docs.google.com/forms/d/e/1F…
Are most AI Red Teams just glorified QA testers?
Coming from classic cyber red teaming, I feel very frustrated about how the term has been adopted in the AI industry.
Curious how many times the definition will drift before the scope of work and responsibilities are finalized?
2
301
Is it Chinese or American research if 10 Chinese PhD students at a U.S. university publish an AI security paper?
1
2
401
Your little weekend project! Help us at OWASP refine the Top 10 for LLM Security.
Share your feedback about the new top 10 candidates here:
docs.google.com/forms/d/e/1F…
The deadline is Sunday, June 30.
Your opinion matters!




