Nischoy launches deterministic prompt injection detector using formal logic

Nischoy has launched a prompt injection detector that uses grammar-based language analysis and SMT-solver-checked formal logic policies instead of an LLM, running on an ordinary CPU without a GPU. Because each verdict traces back to the extracted action and target plus the exact policies that fired, the same input always produces the same answer, the author says. In the author's local benchmark — an in-process detector on an Apple M4 CPU with 120 warm requests and network excluded — a typical check had a median latency of 4.7 ms (p95 6.0 ms), with the first request taking about 101 ms. The author states this is not a production latency measurement. The detector can be tried at nischoy.ai.

Generated from reports · updated 1 h ago

Not enough continuous observations yet to draw a trend.