SECURITY Signal 436
Agentic AI crossing system boundaries escalates security risks
↗
In a podcast, Tracy Bannon discusses how agentic AI crossing system boundaries escalates security risks, and how the SDLC faces novel challenges from non-deterministic agents and cognitive overload on developers.
For engineers building with AI agents, the key risk is not the agents themselves but the boundaries between systems they cross, which expand attack surfaces. The traditional SDLC, designed for human developers, does not account for non-deterministic agents that change behavior during testing, so teams must rethink testing and review processes. Overreliance on AI also risks diminishing human skills in complex domains.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Security risks escalate when agents cross system boundaries between different software ecosystems.
The SDLC was designed for humans; agentic development introduces non-deterministic outputs and testing challenges.
Overreliance on AI leads to diminishing human skills and cognitive overload from reviewing AI-generated code.
THE CLUSTER