Tuesday, July 21, 2026

1. Regulation & Policy

   Washington bars AI as the sole basis for prior-authorization denials and requires reviewer credentials in denial notices

   Washington SB 5395, in force since June 11, 2026, permits carriers to use AI to process and approve prior authorization but requires that only a licensed physician or licensed health professional acting within scope may issue a medical-necessity denial, and obliges carriers to disclose the credentials, board certifications, and specialty of whoever had clinical oversight — a disclosure that gives post-acute providers a concrete basis to test whether a denying reviewer was qualified for the service denied.

   HighBy Washington State SenateRead at source ↗

2. Security & AI Risk

   Health-ISAC and Quest Diagnostics assess autonomous vulnerability-discovery AI as a systemic health-sector risk

   A TLP:WHITE threat bulletin published May 12, 2026 examines a limited-preview model with autonomous zero-day discovery and weaponization capability, warning that misuse risk is comparable to Cobalt Strike or Brute Ratel, that competing firms may match the capability within six to twelve months, and that unregulated equivalents are expected to proliferate through unrestricted or open-source release by mid-to-late 2026 — a forecast the July agentic intrusion now illustrates.

   MedBy Health-ISACRead at source ↗

3. Security & AI Risk

   HSCC and Health-ISAC run "Operation Vital Signs" national health cyber exercise July 21-22

   The two-half-day exercise tests enterprise and cross-sector response and recovery including impacts to critical functions and patient safety, and lands alongside a previewed 2026 HSCC AI cybersecurity guidance series covering education and enablement, cyber operations and defense, governance, secure-by-design principles, and third-party risk and supply-chain transparency.

   MedBy Industrial CyberRead at source ↗

4. Security & AI Risk

   Safety guardrails on hosted models blocked breach forensics, forcing a self-hosted open-weight fallback

   Incident responders found that commercial frontier-model APIs refused to process real attack commands, exploit payloads, and command-and-control artifacts because guardrails cannot distinguish a defender from an attacker, so the forensic analysis ran on an open-weight model on internal infrastructure — a gap healthcare incident-response plans should close by pre-vetting a self-hostable analysis model before an incident, which also keeps attacker data and credentials inside the environment.

   HighBy Hugging FaceRead at source ↗

Get the Daily Dispatch by email · free, about 8 a.m. Eastern · RSS