Crashing Drones and Hijacked Cameras: CyberTrust Meets CyberPhysical
summary
Microsoft Research describes a drone verification stack that embeds a Bayesian classifier into probabilistic signal temporal logic, producing a probabilistic safety program that takes a planner, classifiers and specs and either proves every run safe or returns a counterexample; that approach is state of the art. An ambitious effort called Everest verified HTTPS aims to produce a fully verified, drop‑in HTTPS replacement (work in progress). The SEAL homomorphic library supports encrypted ML demos—4096 MNIST predictions in five minutes and a pneumonia risk prototype—and is publicly released.
AI-written summary · about
chapters (18)
- 0:00Start
- 3:56What Could Go Wrong
- 4:44Cyber Trust for Cyber Physical Systems
- 5:12Research Challenges
- 6:10Reliability
- 6:28Technical Challenges
- 8:58Andre Plotzer at Carnegie Mellon on Differential Dynamic Logic
- 12:23Safe Cyber Physical Systems
- 19:25Security
- 28:21How Can We Trust the Cloud
- 28:32Trusted Cloud
- 30:51Invasion of Privacy
- 31:35The Data Life Cycle
- 32:14How Can We Trust the the Data Holders with Our Data
- 33:29Privacy Technologies
- 35:25Differential Privacy
- 35:58The Use of Cryptographic Cryptography
- 44:48Questions
related talks
25mTrust Management in Digital Ecosystems
19mTransform CyberSec Governance with GenAI
17mPreserving integrity and trust in AI
23mTrust, Track, and Verify: Securing AI Pipelines End-to-End
10mZero Trust Hardware Design for Embedded Systems
24mFrom Trust Assumptions To Trust Evidence: Why PKI and Confidential Computing
40mBuilding Trust in Supply Chains
36mSecurity for AI and Multi-Party Collaboration with Confidential Computing (Oh, and...