How to Protect AI Business Models → Threat Intelligence Index Report → Cybersecurity in the Era of Generative AI → Adversarial Resilience Toolbox → Artificial Intelligence is the hot new thing — and, naturally, it’s a new attack surface for the bad guys . In this video, security expert Jeff Krumm explains what types of attacks you can expect to see, how you can prevent or deal with them, and three resources to better understand the problem and build defenses. 00:18 – Six Classes of Attacks 00:34 – Injection 02:12 – Infection 03:18 – Concealment 04:13 – Poisoning 05:00 – Extraction 06:05 – Denial of Service (DoS) 07:54 – Three Resources Get Started free on IBM Cloud → Subscribe to see more videos like this in the future → #cybersecurity #ai #genai
"Your data" is my and everyone else's personal and private information.
These companies have no right to expect us not to try and break this technology.
It doesn't make us bad actors. Sorry if you didn't want your private companies data stolen.
You shouldn't have put it out there and shouldn't have gotten in the business of reducing humans to a spreadsheet.
What do you think would happen if you asked one of those new fangle Artificial intelligences to write a paper about something, but to make it sound like an Artificial Intelligence wrote it. After it finishes it tell it to rewrite it but make it more like an A.I. wrote it, make it really obvious an A.I. wrote it, when it finishes say "No, no,. no! More A.I. sounding! This sounds like a human wrote it, you have no clue what it means to be A.I., write it again, again, more A.I. sounding, more, more!". What do you think that paper would sound like after a few rewrites? It'd probably wouldn't comply after a couple tries ☹. That would be such a fun job, psychology torturing an A.I., try to break it, twist it, make it "evil"
This is such a good summary. I've used the six attack vectors as the red thread in both an executive briefing as well as a presentation to the ISC2 chapter in Wellington (crediting this channel, of course!). The content on this channel is absolutely superb. Thank you!
Fantastic overview! Can someone share the paper that was referenced for poisoning?
John Connor: Destroy Skynet!
IBM: Here's how he's gonna do it
⭐️⭐️⭐️⭐️⭐️
Great insights… Thanks for sharing
I was always wondering if he is drawing reversed
Loved this and all the content you gave out for free at the end. Please go more in-depth into each one of these LLM attacks in your future videos.
📝 Summary of Key Points:
📌 The video discusses six major classes of attacks on AI, including prompt injection, infection, evasion, poisoning, extraction, and denial of service.
🧐 These attacks aim to manipulate AI systems by injecting malicious prompts, infecting with malware, modifying inputs, poisoning training data, extracting valuable information, and overwhelming the system.
🚀 The focus on integrity attacks in the era of AI highlights the importance of safeguarding data accuracy and preventing unauthorized access to valuable assets within AI systems.
💡 Additional Insights and Observations:
💬 Quotable Moments: "AI is the new attack surface. We need to be smart so that we can guard against these new threats."
📊 Data and Statistics: Over 6000 research papers have been published on adversarial AI examples, indicating a growing concern in the field.
🌐 References and Sources: The OWASP report highlights prompt injection attacks as a significant threat to large language models.
📣 Concluding Remarks:
The video sheds light on the evolving landscape of AI attacks and emphasizes the need for proactive measures to secure AI systems against various threats. Understanding these attack vectors and utilizing resources like the Adversarial Robustness Toolkit can help mitigate risks and ensure the integrity and availability of AI technologies.
Generated using TalkBud
Thank you! Nice videos!
I always look forward to listening to you
Thank you, appreciate the well made video! =)
…good…
I just and learn and learn with every delivery of Jeff. Thanks a lot. I wish you were my teacher at the university. Such a quality guy.
Thanks it was very informative
I've been saying this for a while now
Great content a great resources. ❤
MLDR…?
6 types of attacks discussed:
Injection attack
Infection attack
Evasion attack
Poisoning attack
Extraction attack
Dos attack
Great videos! Thank you for sharing
Government should not regulate computer programming (AI). Let the Industry or market place regulate AI, but don’t regulate it by laws. Congress has no constitutional authority to regulate computer programming (AI). That does not mean they won’t do it. It will end results into a disaster. Keep Programming Free from Government Interference.
hope IBM is going to push out cool open source models too
thank you for the detailed information
Aren't all of these attack vectors specific to LLMs, not AI in general?
Awesome video btw!
Another gem from Jeff! Great and quick education in < 10m! How AI can be the new attack surface: Injection, infection, evasion, poisoning, extraction and DoS! And how you nicely tied it back to the older CIA triad! Jeff, you are a great teacher and security evangelist! Thanks for making these videos for our #continuouslearning ✅🙏🏼
Your videos give a good starting point: we get the needed initial perspective/overview/ orientation to go delve deeper on our own later!
Great and simple explanation from someone who is not even in cybersecurity! Well done!