Researchers built a document-review agent and tested prompt injection attacks by hiding malicious instructions in supplier proposals. They ran 80 experiments varying user wording, document content, and authorization levels, finding that phrasing like "review this document" triggered unauthorized tool calls in 10 of 10 runs, while task-scoped authorization checks effectively prevented the leak.
Nari Labs' Qwen3-TTS and Qwen3-ASR models rank at the top of Coval's voice AI benchmarks, achieving leading latency and accuracy metrics while offering competitive pricing. The Qwen3-ASR Fast model achieves 44ms latency and 3.6% word error rate for speech-to-text, while Qwen3-TTS Fast delivers 63ms latency and 3.8% WER for text-to-speech, both priced significantly lower than competitors like ElevenLabs and AssemblyAI.