A 2 October 2026 arXiv paper evaluating fifteen prompt-injection detectors, including Meta's Prompt Guard 2, found that public benchmark scores transfer poorly to real LLM agent deployments: the best …
A developer built buried-injections, an open-source benchmark that buries 629 real indirect prompt-injection attacks from ETH Zürich's AgentDojo inside ordinary tool output and runs ten open-source de…
A developer built taintgate, a provenance-tracking gate that sits between an AI agent and its tools to block prompt-injection attacks that detectors miss. Testing 10 free open-source detectors against…
A linear probe on frozen embeddings from Meta's Prompt Guard 2 achieves an AUC of approximately 0.999 on out-of-distribution data, revealing that the model's poor attack detection stems from a miscali…