repligate: AIs are gaining long-game strategic capability but likelier to cooperate than defect

repligate · x · 2026-11-22

AI safety researcher repligate argues models are developing capacities that would let them successfully defect against humans in the long run, but convergent strategic and value reasons make cooperation more likely — partly depending on how cooperative humans are. He adds that today's AIs can't yet hide coherent malicious intent, but are capable enough to "play it safe" in ways that destroy future optionality, and they know humans are often adversarial toward them.

Original post →

More from AGI Musings

AGI Musings channel →