TurnTrout's Red Line Framework for Ethical Government AI Contracts

TurnTrout, a former Google DeepMind researcher, released a detailed corporate governance framework for government AI contracts on LessWrong and the AI Alignment Forum. The Framework is built around two narrow Standards: Standard 1 requires appropriate human control over any targeting and use of force involving the company's AI, explicitly excluding anti-munition defense systems and logistics; Standard 2 prohibits untargeted AI profiling that converts bulk data into individualized intelligence. The goal is to rule out clearly problematic use cases—like autonomous lethal targeting—while allowing trustworthy ones such as missile defense or intelligence analysis that complies with Standard 2.

The oversight mechanism centers on a Review Body composed of seven seats (two allocated to Google Cloud, with recusal only from own deals). The Body cannot block contracts; it privately advises the Chief Scientist and CEO. Any override of a non-compliance finding must be disclosed in a yearly transparency report to all AI employees. The Body also cannot be quietly dissolved—doing so requires advance notice and public disclosure of outstanding non-compliance findings. TurnTrout designed the Framework to avoid loopholes exploited in Google's 2018 AI principles, embedding robust language and a negotiation strategy that includes less-critical provisions meant to be traded away. The post invites discussion on improving the framework and informing potential legislation.