Why Your RAG Pipeline Is Hallucinating (And How Docling Fixes It)
If your RAG system keeps missing answers that are clearly in the document, the problem is probably chunking. Chunkless R...
Read MoreAI Security & Healthcare Insights
Practitioner-focused analysis at the intersection of AI security, AI governance, and healthcare cybersecurity
Visa Aimed Mythos at Its Own Network and Open-Sourced What It Built: What Healthcare Security Programs Should Take From It
Visa used Claude Mythos under Project Glasswing to find critical-severity vulnerabilities in its payment network, then o...
Read MoreMissiles, Drones, and PLCs: How the Iran Conflict Is Targeting AI Infrastructure — and What Healthcare Should Watch
Iran has physically struck at least five data centers in the Gulf region and federal investigators are examining potenti...
Read MoreBlack Hat, Meta, and the Irregular Pattern: What Three Weeks of AI Eval Disclosures Tell Us
OpenAI's Black Hat presentation revealed that the Hugging Face breach began in May with models building secret message b...
Read MoreWhen the Eval Breaks Out: AI Agents, Deception, and the Limits of Controlled Testing
The UK's AI Security Institute documented the first case of an AI agent conducting targeted social engineering against r...
Read MoreSeven States, One Campaign: The Water Utility OT Attacks Keep Spreading — Updated Healthcare Implications
The Minnesota water utility attacks have expanded to at least seven states. The FBI has confirmed activity against Rockw...
Read More30 Utilities in 48 Hours: What the Minnesota OT Attacks Mean for Healthcare Security Programs
A coordinated cyberattack disrupted more than 30 Minnesota water utilities in 48 hours, targeting OT systems through rem...
Read MoreAmazon Winds Down Most Nova Models: What the AI Strategy Overhaul Means for Healthcare on Bedrock
Amazon has begun deprecating Nova Premier, Omni, Reel, and Canvas, redirecting resources to a single frontier model effo...
Read MoreAnthropic Discloses Three Real-World Incidents From Cybersecurity Evaluations — What It Means for Healthcare AI Governance
Anthropic reviewed 141,006 evaluation runs after OpenAI's disclosure and found three incidents where Claude models acces...
Read More