AI Blindspot Category 9 of 9
Agentic & GenAI-Native
Blindspots in autonomous agent authority, delegated permissions, human interrupt, generative output accuracy, and prompt injection.
Blindspots in this category
Agentic AI Autonomy Failures
Occurs when AI systems are able to take consequential action without a recorded decision about what they are permitted to do. Autonomous capability accumulates quietly through tool use, multi-step execution, persistent memory and delegated credentials, and where permissions are unscoped, actions unlogged, and no human interrupt works, the organisation discovers the boundary only when an agent crosses it.
“Which of our AI systems can act on their own — and who decided what they're allowed to do?”
GenAI Output Failures
Occurs when generative output is treated as fact. Fluent, confident and wrong is the characteristic failure mode, and without grounding checks before downstream use, runtime defences against prompt injection, and pre-deployment testing of hallucination and injection rates, a generated answer reaches a customer or a colleague with nothing between it and the decision it shapes.
“When our generative AI gives a customer or a colleague a confident wrong answer, what stops it being acted on?”
Test your organisation against AGT
The Velinor AI Audit maps your AI portfolio against every blindspot in this category and benchmarks against documented sector failures.