"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis

The Future of AI Security with Adam Wenchel, CEO of Arthur.ai

22 snips
Sep 19, 2023
Adam Wenchel, CEO of Arthur.ai and pioneer in AI security, discusses the evolving threats in AI, especially regarding large language models. He delves into historical cybersecurity challenges and reveals the vulnerabilities of LLMs to adversarial attacks. Wenchel emphasizes the critical balance between fraud prevention and customer satisfaction in AI models. He also introduces advanced monitoring tools like Arthur Shield and Arthur Bench to improve governance and reduce risks like AI hallucinations, spotlighting the importance of compliance in the financial sector.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

AI Security vs. Cybersecurity

  • AI security differs from traditional cybersecurity because AI models are probabilistic, not deterministic.
  • Their behavior isn't explicitly coded, requiring a shift in security mindset.
ANECDOTE

Boundary Detection Attack

  • Fraudsters exploited a bank's AI system by keeping luxury purchases under $500.
  • This "boundary detection attack" highlights how attackers probe AI models for weaknesses.
ADVICE

Mitigating AI Risks

  • Implement real-time observability for AI models to detect unusual patterns quickly.
  • Develop robust training routines to make models less susceptible to attacks.
Get the Snipd Podcast app to discover more snips from this episode
Get the app