[bearer_token]

Is this a sufficient risk that an AI company would be wise to hire an adversarial security engineer? Or is this still well below hypothetical?

[WrtCdEvrydy]

Depends.

Isn't an adversarial security engineer just a regular QA guy for your model?

An application expecting a number between 10 and 20 and receiving a number larger than an integer causing an error isn't a new issue.