The Hidden Cybersecurity Risks When Doctors Use AI Diagnostics | Ep. 58 скачать в хорошем качестве

The Hidden Cybersecurity Risks When Doctors Use AI Diagnostics | Ep. 58

About 10 years ago, Singapore experienced a national cyberattack. Patient health records were exfiltrated from the healthcare system. That breach forced the government to rethink how medical devices connect to hospital networks and what security standards manufacturers should meet. The result is a four-level Cybersecurity Labeling Scheme for medical devices. CLS-MD gives hospitals transparency about which devices have undergone rigorous security testing and which meet only bare minimum requirements. Level one represents basic compliance. Level four requires enhanced security protections and independent third-party code review. The timing matters because two massive trends are colliding. First, clinicians are using unauthorized AI tools. Nearly a quarter of doctors are uploading X-rays to ChatGPT, asking AI to diagnose patients, doing it without any organizational oversight or security controls. Second, AI is being used to write medical device code. Fast. Efficient. But potentially vulnerable if not guided correctly. Both trends create cybersecurity risks that didn't exist five years ago. Adversarial attacks where someone manipulates a few pixels in a medical image to trigger a false diagnosis. Data poisoning where corrupted data gets introduced into training sets. Model drift where AI performance degrades over time. And automation bias where people trust AI more than humans saying the exact same thing. This episode walks through how adversarial attacks work, why automation bias is such a dangerous problem in clinical settings, how to use AI safely for medical device development, and what Singapore's four-level labeling scheme actually requires. It also covers AI scribes, dialect challenges in Southeast Asia, and why designing systems that fail safely is more important than designing systems that never fail. If you're building AI-powered medical devices or digital health tools, this conversation has a lot of practical value. Episode Breakdown: 00:01 Welcome and intro 00:31 Guest background: military, AI, healthcare 01:09 Singapore's mandatory military service 03:09 Emerging AI threats in healthcare 03:45 The 23% problem: unauthorized AI tool usage 04:40 Clinicians uploading X-rays to ChatGPT 05:43 Adversarial attacks: manipulating pixels to trigger false diagnoses 08:15 Data poisoning and model drift in medical AI 11:30 AI hallucinations and diagnostic accuracy 14:20 Using AI for medical device software development 17:45 Security vulnerabilities in AI-generated code 20:30 AI as a pair programmer: how to do it right 23:15 Building guardrails for AI failures 26:40 Automation bias: why people trust AI over humans 28:50 AI scribes: what gets included and what gets left out 31:20 Dialect challenges for AI in Southeast Asia 34:05 AI pre-triage systems in Singapore clinics 36:32 Real example: ChatGPT inventing pricing for services 37:25 Designing AI to fail safely 37:40 Key point: Human interpretation is still essential The Med Device Cyber Podcast is brought to you by Blue Goat Cyber, cybersecurity experts providing essential security solutions for the medical device industry. Learn more by visiting https://bluegoatcyber.com. If you're interested in our services or partnering with us, schedule a Discovery Session: https://meetings.hubspot.com/blue-goa... Christian Espinosa is the CEO and Founder of Blue Goat Cyber. Trevor Slattery is the Chief Operating Officer at Blue Goat Cyber. Christian Espinosa on LinkedIn:   / christianespinosa   Trevor Slattery on LinkedIn:   / trevor-slattery-34852b1a9   Blue Goat Cyber on LinkedIn:   / blue-goat-cyber   Blue Goat Cyber on Instagram:   / bluegoatcyber   Blue Goat Cyber on Facebook:   / bluegoatcyber   Blue Goat Cyber on YouTube:    / @bluegoatcyber