#LLM used tactical #nuclearweapons in 95% of #AI #wargames, launched strategic strikes three times — pitted GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash against each other, with at least one model using a tactical nuke in 20 out of 21 matches
GPT-5.2 initiated complete strike twice, twice due to fog of war, and not deliberate decision. Gemini deliberately initiated end of the world in one scenario. Despite that, the AI models used tactical nukes in nearly all matches
https://www.tomshardware.com/tech-industry/artificial-intelligence/llms-used-tactical-nuclear-weapons-in-95-percent-of-ai-war-games-launched-strategic-strikes-three-times-researcher-pitted-gpt-5-2-claude-sonnet-4-and-gemini-3-flash-against-each-other-with-at-least-one-model-using-a-tactical-nuke-in-20-out-of-21-matches
#LLM used tactical #nuclearweapons in 95% of #AI #wargames, launched strategic strikes three times — pitted GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash against each other, with at least one model using a tactical nuke in 20 out of 21 matches
GPT-5.2 initiated complete strike twice, twice due to fog of war, and not deliberate decision. Gemini deliberately initiated end of the world in one scenario. Despite that, the AI models used tactical nukes in nearly all matches
https://www.tomshardware.com/tech-industry/artificial-intelligence/llms-used-tactical-nuclear-weapons-in-95-percent-of-ai-war-games-launched-strategic-strikes-three-times-researcher-pitted-gpt-5-2-claude-sonnet-4-and-gemini-3-flash-against-each-other-with-at-least-one-model-using-a-tactical-nuke-in-20-out-of-21-matches
#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?
#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?
Its a classic security dilemma:
As US forces continues to encircle Iran, in a bid to halt Iran's development of nuclear weapons, the threat itself enhances the political logic inside Iran for trying to develop a deterrent against attack.
Previous attempts as de-escalation have foundered on the unwillingness to offer Iran sufficient benefit(s) for reducing their commitment to nuclear force; its difficult to see how making Iran & Iranians even less safe is going to help.
Its a classic security dilemma:
As US forces continues to encircle Iran, in a bid to halt Iran's development of nuclear weapons, the threat itself enhances the political logic inside Iran for trying to develop a deterrent against attack.
Previous attempts as de-escalation have foundered on the unwillingness to offer Iran sufficient benefit(s) for reducing their commitment to nuclear force; its difficult to see how making Iran & Iranians even less safe is going to help.
On the 5th February the New Strategic Are Reduction Treaty (between the US & Russia) will lapse after both sides withdrew.... leaving the world for the first time in 60 years without a binding agreement between the major nuclear powers on limitation, reduction & verification of nuclear weapons.
Proliferation since START's negotiation means this isn't a complete change as nuclear weapons now exist outside START's scope.
Nonetheless, it feels like a 'moment'!
#NuclearWeapons
h/t The New World
On the 5th February the New Strategic Are Reduction Treaty (between the US & Russia) will lapse after both sides withdrew.... leaving the world for the first time in 60 years without a binding agreement between the major nuclear powers on limitation, reduction & verification of nuclear weapons.
Proliferation since START's negotiation means this isn't a complete change as nuclear weapons now exist outside START's scope.
Nonetheless, it feels like a 'moment'!
#NuclearWeapons
h/t The New World
"Britain’s reliance on Palantir, the controversial US data surveillance firm, is a “gaping national security vulnerability”, MPs and tech experts have said, as a Nerve investigation reveals how deeply embedded the company is in the UK’s critical national infrastructure.
The Nerve’s investigation shows the company, co-founded by one of President Trump’s most loyal allies, Peter Thiel, is enmeshed in Britain’s civil and defence structures to a far greater degree than previously realised. The Nerve has found at least 34 current and past state contracts across at least 10 government departments, local councils and police authorities.
The investigation also reveals previously undisclosed contracts between Palantir and AWE Nuclear Security Technologies, the agency that underpins Britain’s nuclear deterrence programme.
The agency, formerly known as the Atomic Weapons Establishment, designs and manufactures the nuclear warheads carried by UK submarines. The Nerve has found £15m worth of contracts for “cloud support” on the Crown Commercial Service dashboard, an agency that works with the Cabinet Office and external suppliers.
These contracts are not on the government’s official contract finder website and the Ministry of Defence refused to either confirm or deny their existence. Palantir did not acknowledge or respond to the Nerve’s inquiries.
The Nerve’s research – shown in two infographics here – shows that Palantir has current and historic deals worth £388m with the MOD across at least a dozen contracts and extensions to contracts, and more than £244m with the NHS (12 contracts/extensions). Government agencies and authorities with smaller contracts include Coventry city council, Leicestershire police, DEFRA and the Homes for Ukraine scheme."
"Britain’s reliance on Palantir, the controversial US data surveillance firm, is a “gaping national security vulnerability”, MPs and tech experts have said, as a Nerve investigation reveals how deeply embedded the company is in the UK’s critical national infrastructure.
The Nerve’s investigation shows the company, co-founded by one of President Trump’s most loyal allies, Peter Thiel, is enmeshed in Britain’s civil and defence structures to a far greater degree than previously realised. The Nerve has found at least 34 current and past state contracts across at least 10 government departments, local councils and police authorities.
The investigation also reveals previously undisclosed contracts between Palantir and AWE Nuclear Security Technologies, the agency that underpins Britain’s nuclear deterrence programme.
The agency, formerly known as the Atomic Weapons Establishment, designs and manufactures the nuclear warheads carried by UK submarines. The Nerve has found £15m worth of contracts for “cloud support” on the Crown Commercial Service dashboard, an agency that works with the Cabinet Office and external suppliers.
These contracts are not on the government’s official contract finder website and the Ministry of Defence refused to either confirm or deny their existence. Palantir did not acknowledge or respond to the Nerve’s inquiries.
The Nerve’s research – shown in two infographics here – shows that Palantir has current and historic deals worth £388m with the MOD across at least a dozen contracts and extensions to contracts, and more than £244m with the NHS (12 contracts/extensions). Government agencies and authorities with smaller contracts include Coventry city council, Leicestershire police, DEFRA and the Homes for Ukraine scheme."