I sympathize with this dilemma. It's a real-life version of the Trolley Problem[1].
Or imagine an idle threat is made off-hand in a private gmail, so pre-emptively disclose that threat to law enforcement?
In the sci-fi Minority Report scenario, Google could use AI language models on all gmail to make a judgment call on "emergent danger of serious physical injury". If I plan to go bunjee jumping with an unreliable provider, Google scans my gmail and tells law enforcement?!?
I know a person who was banned from twitter and was nearly fired because he dead-named somebody. The explanation was that it was literal violence against the person he dead-named, because so many trans people self harm if they don't feel that their identity is affirmed. The person who was banned says it was a force of habit because the person had only very recently went public with their new identity, but it did not stop the ban hammer. He almost lost his job as well because the tweet was posted in a public slack channel. HR left it up the person who was dead-named to make the decision, and they decided not to fire him.
I could absolutely see a scenario like the above happening in gmail or some other Google product, and the decision-maker deciding that it was a threat to a person's life and should be disclosed. To some people that makes perfect sense. To others it does not. The point is simply that when we let humans make subjective decisions like that, we get all the downsides of human judgment.