“Peter Steinberger asserts that weaker, cheaper, or local AI models are significantly more susceptible to prompt injection attacks than more advanced models.”