Nick Routledge@nick_routledge
Reply to @nick_routledge

@NoemaMag @AndrewSorota 21/ That's the shift: AI-mediated NC3 collapses perception and execution into one loop operating at machine speed. The stack references itself—at a tempo, opacity, and unpredictability lashed to doomsday incentives—pointing the result at launch authority.

Image and subtweet here: 1/2 "Even in neutral scenarios without initially provided conflicts, Big Tech's AI models chose violence and escalated to nuclear strikes in simulated wargames. All models show signs of sudden and hard-to-predict escalations.” NATO Review, 2 April, 2024: "There are myriad roles for AI in our nuclear defences ...Concern over AI should not preclude the use of AI in strengthening nuclear deterrence ...There is, of course, the potential for an arms race in AI-based targeting systems for nuclear weapons." Wikipedia entry: Large language models The potential presence of "sleeper agents" within LLM models is another emerging security concern. These are hidden functionalities built into the model that remain dormant until triggered by a specific event or condition. Upon activation, the LLM deviates from its expected behavior to make insecure actions.
Image description

Original X alt text

Image and subtweet here: 1/2 "Even in neutral scenarios without initially provided conflicts, Big Tech's AI models chose violence and escalated to nuclear strikes in simulated wargames. All models show signs of sudden and hard-to-predict escalations.” NATO Review, 2 April, 2024: "There are myriad roles for AI in our nuclear defences ...Concern over AI should not preclude the use of AI in strengthening nuclear deterrence ...There is, of course, the potential for an arms race in AI-based targeting systems for nuclear weapons." Wikipedia entry: Large language models The potential presence of "sleeper agents" within LLM models is another emerging security concern. These are hidden functionalities built into the model that remain dormant until triggered by a specific event or condition. Upon activation, the LLM deviates from its expected behavior to make insecure actions.