MAI-Transcribe-2-Streaming posts a 2.5% final word error rate at 0.13 seconds, per Microsoft, a combination that Deepgram and Cartesia can't match at once. Compare the latency, accuracy and intro pricing.

A user's Claude message threatening a Florida sheriff's office went from Anthropic's safety filters to a human reviewer to a felony arrest, per TechSpot. Read what Anthropic's emergency-disclosure policy allows.

Invalid AI-generated reports swamped triage, per TechCrunch, so Google's OSS VRP closed to new product-vulnerability reports October 1 and promises only a Q1 2027 update. See which reports still get through.