tszzl predicts open-weight models will be banned after a major AI disaster, sparking e/acc backlash

Prominent AI researcher tszzl (a16z advisor, author of Echo/Twin) predicted that after some major AI disaster, open-weight models would be banned 'before long.' He clarified this is a prediction, not a preference: he loves open models, is running a Kimi K3 finetune on Tinker, and hopes Kimi, DeepSeek and others keep building—just behind monitored APIs rather than open weights. The thread was amplified by Sentdex and entangled with Roon's regulatory capital plans.

Confirmed

tszzl's core argument: as open models approach superintelligence, 'offense dominates defense' in the safety game, and free distribution could enable bad outcomes; both the US and China would then try to control them. This is unrelated to the 'brake on the frontier' debate—he neither wants to nor could ban open source.
He said he cannot imagine a world a year from now where open models are not tightly regulated, and would happily admit being wrong if open models prove harmless and unregulated.
He sincerely hopes for a flourishing of countless kinds of minds, but says the outcome depends on how much chaos open models cause.
jdpressman added that the judgment is less about open models' danger than about AI deployment structurally favoring scale, making tighter regulation nearly inevitable.
e/acc founder Guillaume Verdon (@beffjezos) condemned regulators seeking to ban open source and strip citizens of the ability to own and control their AI hardware and weights, calling it 'pure evil disguised as safety benefits.'
Matthew Berman argued that forcing open models behind monitored, restricted APIs concentrates power and is bad for the world overall.

Why it matters

The dispute touches the core divide in AI governance: one side sees tightening control of open weights after a major incident as nearly inevitable given superintelligent capabilities and deployment economics; the other, led by e/acc, views it as an assault on personal freedom and decentralized innovation. tszzl offered a testable one-year prediction, while opponents counter that restrictions bring power concentration rather than safety—an opposition that will keep shaping the open-source community and regulatory trends.

2026-09-12 ~ 2026-09-13 · 10 related posts

Primary sources

tszzl: I love open models, but offense dominates defense as they approach superintelligence — tszzl · 2026-09-13 14:21 (UTC+08:00)
tszzl follows up: my prediction isn't a preference, I'm not trying to ban open source — tszzl · 2026-09-13 14:21 (UTC+08:00)
e/acc Founder Beff Jezos Blasts Open-Source AI Bans as 'Pure Evil Disguised as Safety' — beffjezos · 2026-09-12 08:11 (UTC+08:00)

[source] e/acc Founder Beff Jezos Blasts Open-Source AI Bans as 'Pure Evil Disguised as Safety' — beffjezos · 2026-09-12
tszzl predicts open-source models will be banned after a major AI disaster; Beff pushes back — beffjezos · 2026-09-13
tszzl predicts open-source AI will be banned after a major disaster, wants monitored APIs — mimi10v3 · 2026-09-13
Open-source AI debate: forced API monitoring is a power-concentrating outcome — chris_j_paxton · 2026-09-13
Open-weight models likely face harsh regulation within a year, argue researchers — scale beats safety logic — teortaxesTex · 2026-09-13
[source] tszzl: I love open models, but offense dominates defense as they approach superintelligence — tszzl · 2026-09-13
[source] tszzl follows up: my prediction isn't a preference, I'm not trying to ban open source — tszzl · 2026-09-13
tszzl: I hope a thousand minds can bloom — the outcome depends on how much chaos ensues — tszzl · 2026-09-13
tszzl warns open models nearing superintelligence: offense dominates defense, US and China will intervene — JasonBotterill · 2026-09-13
Researcher predicts open-source AI models will be banned after a major disaster — TinfoilTricorn · 2026-09-13