#AI can’t stop recommending nuclear strikes in #wargame simulations
Leading AIs from #OpenAI, #Anthropic and #Google opted to use #nuclearweapons in simulated war games in 95% of cases
The scenarios involved intense international standoffs, including border disputes, competition for scarce resources and existential threats to regime survival.
https://www.newscientist.com/article/2516885-ais-cant-stop-recommending-nuclear-strikes-in-war-game-simulations/

What could go wrong?

AIs can’t stop recommending nuclear strikes in war game simulations

Leading AIs from OpenAI, Anthropic and Google opted to use nuclear weapons in simulated war games in 95 per cent of cases

New Scientist
@BenjaminHCCarr
LLMs haven't been "fed" any Isaac Asimov - or perhaps they don't recognise that they are robots. #3Laws
@Nazani @BenjaminHCCarr but Terminator was also in their training set
@inecas @BenjaminHCCarr No doubt... and we're going to need more to beat this than a woman who can slam-cock a Rem 870.

@BenjaminHCCarr @hacks4pancakes

Maybe training LLMs on text from a violent culture was not a great idea.

@BenjaminHCCarr There was literally a movie about this back in the day, right?
@BenjaminHCCarr maybe calling llms and their ilk 'ai' is part of the problem. Because of that label, people believe that they think, and compare them to ai characters portrayed in stories. The use of such terminology causes the llm to reproduce tokens related to 'ai' from their learning corpus stolen from hundreds of sci fi stories.
Of popular fiction had their ai characters commonly playing hopscothc, that action is what they would be outputting.
@BenjaminHCCarr Why oh why am I not--for even a nanosecond--surprised?
@BenjaminHCCarr Super subtle Skynet.
Anthropic ditches its core safety promise in the middle of an AI red line fight with the Pentagon

Anthropic, a company founded by OpenAI exiles worried about the dangers of AI, is loosening its core safety principle in response to competition.

CNN

@BenjaminHCCarr

https://www.independent.co.uk/news/world/middle-east/saudi-arabia-trump-uranium-deal-b2925502.html

What could go wrong with gamification of extinction level events?

Same thing that happened with flight simulators that permitted gamers to simulate flying into the World Trade Towers.
https://en.wikipedia.org/wiki/Microsoft_Flight_Simulator_2002

https://en.wikipedia.org/wiki/Zacarias_Moussaoui

Remember, the Saudis funded 9/11, Trump gave them nuclear technology, & the Saudis own Electronic Arts now.

https://en.wikipedia.org/wiki/Alleged_Saudi_role_in_the_September_11_attacks

https://www.axios.com/2025/11/21/911-families-lawsuit-links-saudi-officials-to-plot-despite-princes-denial

1/

Experts warn Saudi Arabia may have uranium enrichment under proposed deal with US

Saudi-US proposal comes amid Iran tensions

The Independent
@BenjaminHCCarr Did nome of these people go to the cinema in the 80s and 90s?
@BenjaminHCCarr
Well…that might make it easier for the #AI to take over…
@BenjaminHCCarr how about a nice game of chess?
@BenjaminHCCarr @ShaulaEvans Those of us who grew up in the 80's probably don't find this even vaguely surprising.
"Would you like to play a game?" has been the defining AI phrase for me since WarGames came out in 1983