You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
deathbench.com LLM's causing death directly is no longer science fiction.
There are multiple documented cases of an AI either pushing someone towards suicide or directly helping plan violent crimes.
As far as I know, there is no current well-maintained repository of these instances, especially one that openly publishes methodology in a way that is an evergreen resource.
For advocacy orgs and policy orgs, it seems extremely important to have a resource to point to that maintains a cumulative index, rather than AI labs being able to run a PR playbook on every single instance in isolation.
I want to provide an easily searchable, easily fact-checkable public index of all deaths that are caused by AI to the best of our ability. I think that the best way to convince people of real-world impact before it is too late is to show it publicly.
It seems important to have clear, conservative, easily understood methodology, in order to avoid getting dismissed easily by PR teams and apologists.
I want to start building a reputation of trustworthiness so that by the time we have our first AI mass casualty event, this will be an established resource that people can go to see harms.
hiring someone to help me gather original data. Right now crucial details like original chat transcripts are missing from most cases. These are typically only in the hands of the family & the frontier lab, (along with lawyers, likely under attorney/client privilege). Reaching out to the family delicately, personally, and explaining that the full info (even if some private details are redacted) may save others is critical.
funding for myself. realistically projects that are not free tend to get more work and love.
spreading the word. I would like help with distribution of this project. I'm not exactly sure how I will spread the word here, but in-person events in San Francisco are probably most likely.
For now, this is a solo project. I am a two-time startup founder (last project was happycampr.ai, techstars '25). Currently working in sf at lore.link as my day job.
I've been working in AI for a while, but this should be my first project of this type.
In the past I led a (for-profit) group to teach kids how to code in my home state of Louisiana. We ended up changing state policy to make it easier for kids to code. I had to push through an initiative in partnership with the Louisiana Department of Corrections, so I have some policy experience.
Anecdotally, when I talk to people about this, very few people are aware that AI systems have already taken action, likely causing the death of people in the real world. I expect that if this fails, that lack of awareness will continue to a greater extent.
Nothing for this - launched from personal funds/time.
Footnote: Currently, I am planning this as an information-only project. I am not ruling out the possibility of future lobbying.
There are no bids on this project.