Forward Focus: Perspectives on AI, Hype, and Security
summary
Large language models show real capabilities but lack open benchmarks, so claims are mixed: companies must register foundational models under the EU AI Act and perform assessments; ML components behave like software that needs patching (PyTorch flaws and supply‑chain checks); research shows protein‑folding transformers enable faster drug design; off‑the‑shelf malicious tools report only a 27% success rate in experiments; defenders and attackers are in a red queen race as agents (LLMs with tool access) start appearing.
AI-written summary · about
chapters (13)
- 0:00Introduction and context
- 3:51Evaluating AI capabilities
- 7:20Security for AI products
- 10:48Emergence and testing
- 12:18Offensive security in AI
- 14:01Security and regulation
- 15:46Compliance and accountability
- 17:37Cross-functional cooperation
- 22:29AI in malicious attacks
- 25:29Adversarial adoption
- 27:33Research and future outlook
- 29:58Learning and best practices
- 34:13Final thoughts and predictions
related talks
42mForward Focus – The Impact of Artificial Intelligence
30mIntroduction to AI Safety. Let's be ready before it's too late!
45mSecuring AI: There Is No Try, Only Do!
49mHype and Reality: How to Evaluate AI/ML in Cybersecurity
24mSoftware 3.0 and the AI Engineering Landscape
32mBeyond the Buzz: Navigating AIs Myths and Realities
43mNavigating the AI Frontier: Investing in AI in the Evolving Cyber...
01h 24mOffensive Cybersecurity by AI Promises and Pitfalls