zeshen comments on Don’t design agents which exploit adversarial inputs

[zeshen]

Thanks for the explanation!