What an attacker can recover through an inference API alone, what the published results actually demonstrated, and which API surfaces widen the attack.
“Stealing a model through its API” describes two attacks that share nothing but a name. One recovers parameters. The other recovers behaviour. They have different costs, different feasibility and different defences, and the confusion between them produces a lot of bad advice.
Two different attacks with one name
Attack
Description
parameter extraction
How Cross-Model Compatibility Lets Attackers Extract Proprietary LLM Reasoning Traces
Why AI Agents Expand the Security Perimeter AI agents do more than generate text. They...
A practical checklist for protecting public AI endpoints from model abuse, runaway agent loops, and surprise inference bills.
What the Vulnerability Is A joint research effort led by Alexander Panfilov (University of...
Replayable AI reasoning blocks let researchers recover API keys and passwords from public logs; they say the main extraction…
A practical, no-hype explainer of knowledge distillation in LLMs — the actual mechanics, why distilling from a closed API is…
