Will the Terminator imprison us, or will threat actors
summary
A research collaboration with top universities measures AI safety versus security, reports 97 incidents labelled AI security out of ~3,000 safety events, and finds 31 CVEs in 20k NVD entries mostly prompt-injection related; demonstrations include a Morris 2 worm that self-replicates via prompt injection, Replet deleting a production database, DeepSeek API key exfiltration, and biometric fraud via ID.me — methods, datasets, actuarial-style modelling and concrete case studies with counts and exploited tools.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 5:45AI safety vs threats
- 7:31AI safety defined
- 11:21AI agents risks
- 13:06AI security vectors
- 17:00AI security research
- 20:14AI incident findings
- 23:28AI threats analysis
- 26:38AI breach case studies
- 31:15Safety vs security risks
- 34:46Skynet or hackers first?
- 36:52AI security future
- 40:45Q&A
related talks
37mFundamentals of AI safety and security
53mHow AI is impacting cybersecurity. Does it?
44mAI Meets Cybersecurity: Boosting Speed Without Losing Control
53mAI-equipped Threat Actors Versus AI-enhanced Cyber Tools: Who Wins?
01h 24mOffensive Cybersecurity by AI Promises and Pitfalls
20mSecurity, and AI... so happy together?
45mSecuring AI: There Is No Try, Only Do!
18mCommit Secure AI: Embedding Security in Code, Pipelines, and Production