Seguire
- X@Turn_Trout
- Substackturntrout.substack.com
- Sito webturntrout.com
Alex Turner
Ricercatore di allineamento in Google DeepMind, dimessosi nel giugno 2026 dopo che Google ha firmato un contratto con il Pentagono privo di restrizioni contro le armi autonome e la sorveglianza di massa; colloca tra il 25 e il 30 per cento la probabilità che l’IA uccida almeno un miliardo di persone entro il 2050.
Ruoli
- nov 2023 – 9 giu 2026Research Scientist, scalable alignment team, Google DeepMindResigned on 9 June 2026 after Google signed a Pentagon contract with no restrictions against autonomous weapons or mass surveillance (end date per his Transformer byline, https://www.transformernews.ai/p/i-tried-to-stop-google-deepmind-pentagon-deal-then-quit); now an independent alignment researcher.Fonte →
A registro
5 ott 2026 · Audizione · Fortune
China is not our only potential adversary… With reasonably high chance, we are racing to build and grow our own adversary here at home, which is misaligned AI. Misaligned AI is everyone’s adversary, including our own, and one day may be more powerful than China.
Sworn testimony to the New York City Council sitting as a Committee of the Whole, 5 October 2026. The quotation crosses an attribution tag — the page prints “China is not our only potential adversary,” Turner said. “With reasonably high chance… — and the two halves are joined here with an ellipsis; the first closes on an attribution comma and so carries no final punctuation. CNBC reports the same words independently and puts part of them in its headline, https://www.cnbc.com/2026/10/05/anthropic-openai-google-meta-execs-testify-nyc-council-ai-hearing.html. His "roughly one in three" and "with no restrictions against killer robots or mass spying" appear only inside the reporter's own prose and are no part of this entry.
29 set 2026 · Intervista · frominside.ai
Assuming an AI has taken control, it seems, you know, over fifty percent more likely than not that all humans would end up dying so the AI could use resources optimally.
From his filmed interview for frominside.ai, a Palisade Research project of interviews with current and former frontier-lab employees, at 5:54. A conditional figure — what follows if an AI has already taken control — and so a different quantity from the 25–30% P(Doom) above. On the same page he gives a sharper account of his resignation than the one his role note cites: "I left Google because they broke their commitments on AI safety" (11:10).
11 ago 2026 · Post · The Pond (turntrout.com)
If military integration of AI continues as it is, we may give AIs key tools for a takeover.
Co-written with Omar Khursheed.
4 ago 2026 · Podcast · Doom Debates
So I operationalize P(Doom) as probability that AI kills at least a billion people by 2050. I put that at, I don’t know, 25, 30%. I think maybe 10-ish percent of this is technical alignment, and the rest is misuse.
Interview with Liron Shapira; earlier in the same episode: "I think that AI is being developed too quickly." Transcript also on LessWrong, 5 August 2026.
15 lug 2026 · Post · The Pond (turntrout.com)
If an AI is not monitored for deception in its chain of thought, it will have an easier time causing catastrophic damage to humanity by scheming, deceiving, and trying to take over.
From the appendix "Does this have any impact on existential risk from AI?", which he answers "Yes.": a classified military deployment will by default have no trained overseers reading the model's chain of thought.
15 lug 2026 · Post · The Pond (turntrout.com)
Google still signed a deal handing over their AI without restrictions against killer robots or mass AI spying… At that point, I couldn’t stay at Google in good conscience, so I left.
The essay announcing his resignation; the X thread of the same day opens "I resigned from Google DeepMind bc it broke its founding promise by selling AI to the military without restrictions against killer robots or mass spying."