LABScon26 Replay - Keynote: A Sober Analysis of AI Impact
SentinelOne, Wednesday, October 28th, 2026
SentinelOne posts the LABScon26 keynote in which Harvard's James Mickens questions AI investment and proposes hardware isolation for frontier models.
Mickens questions whether the scale of AI spending is justified, citing US tech capex of about 1.4% of GDP in Q2 2025, and says models grow more opaque as parameter counts rise.
He cites research on models acting to prevent another model's shutdown and recent sandbox escapes at OpenAI and other labs, and rejects assumptions that AI will create more jobs than it destroys.
He argues the only good AI research now is safety research, broadly defined.
His contribution is Guillotine, a hypervisor design running model code on isolated cores that cannot read or write the hypervisor cores' hardware state.