Introduced by Randall Degges, Manoj Nair argues for separating software generation from independent security validation. He treats generated code, the execution environment and agent behavior as distinct attack surfaces, all of which can introduce risk even when an individual component looks harmless.
The talk emphasizes inventory, repeatable checks and evidence of real weaknesses. Model-based analysis is paired with deterministic controls rather than treated as a substitute for them. Vendor claims and demonstrations are presented as the speakers' approach, not independent proof of superior performance.
Ezra Tanzer demonstrates how dependencies with no reported vulnerabilities can still differ in maintenance and operational risk. He also examines skills and remote instructions, showing why unchanged local content does not guarantee that externally fetched behavior stays safe.
The closing framework combines governance, observation and response so that security decisions can keep pace with autonomous development. Booth invitations, event promotions and product calls to action are omitted.
Watch on YouTube
