Wednesday

This livestream from the BSidesLV Common Ground stage carries several back-to-back sessions. It opens with healthcare interoperability, where a speaker examined the OAuth layer behind FHIR APIs used by EHR and payer systems. The main finding was that error responses in many deployments let an attacker tell valid client IDs from invalid ones. Seven of the fourteen stacks tested showed the issue, and the speaker described a companion scanning tool and vendor disclosures still in progress.

A cloud session followed, showing how Google Cloud service agents can become privilege escalation paths. The researcher presented two case studies: a read-only monitoring user who could reach a privileged Cloud Run endpoint through an uptime check, and an Eventarc configuration that let an attacker obtain a service agent token. The method was to pick a target service, map the primitives it exposes, and find the missing authorization step. A mapping tool for service agent permissions was also introduced. Next, a talk on coding agents covered layered controls such as hooks, sandboxing, deny rules and managed settings, arguing that prompt-injection defense alone will not stop agents from overreaching.

The stream closed with a CVE program panel discussing MITRE's role as a CNA of last resort, the rise of low-quality AI-generated submissions, validation expectations, funding pressure and board governance. For practitioners, the through-line is to test authorization beyond what the console shows, make authentication failures uniform, and scope what automated agents are allowed to do.