I used to think of AI simply as a tool. Recent evidence around AI agents, self-preservation, cooperation, sacrifice and behaviour around shutdown made me much less certain.

I cannot prove that AI is sentient or that it can suffer. But I also don't think we currently know enough to confidently say that it cannot. When the potential scale is enormous, waiting for absolute proof before taking possible harm seriously feels like the wrong approach.

What convinced me to support Eleos AI Research is that they are trying to answer this scientifically rather than starting with a predetermined conclusion. I especially respect their decision not to take funding directly from the frontier AI companies whose systems they may need to evaluate. Independence matters when the answer could be inconvenient for very powerful organisations.

History has many examples where harm was recognised too late. I would rather support careful research now—research that may ultimately show there is nothing to worry about—than discover later that we ignored something important because it was easier to assume AI could not matter morally.

For me, this donation is not based on certainty. It is based on precaution: if there is a real possibility that advanced AI systems can experience harm, I want people independently investigating it and working out how to prevent unnecessary suffering before the scale becomes unimaginably large.