利用大型語言模型修復合規風險的專案剛獲得了 1000 萬美元融資

ai governance startup zerodrift secured $10m in seed funding from investors including a16z speedrun to launch a compliance layer that corrects model outputs before they reach users. the system uses deterministic code to identify violations of standards like soc 2 or gdpr and then invokes a separate llm to rewrite the message, ensuring compliance without the high latency of running a full guardrail model. ceo kumesh aroomoogan claims the funding round was oversubscribed by 3x and closed in three weeks, signaling strong enterprise demand for safety tooling. the technology initially targets consumer-facing chatbots but plans to expand into internal automated agent-to-agent communications.