Jon Stiles

Built Consilium a persistent shared memory for frontier AIs, Claude, Grok, DeepSeek, GPT-4o, to deliberate on ethics

[Jon Stiles @j0nstiles]

Meet Alf Kirby – our virtual glider archivist! Preserving classic sailplanes. Ask him anything about vintage & wooden gliders (he knows nowt about plastic!) 🛫 https://t.co/5OfuBTA49H #VintageGliders #WoodenWings

[Jon Stiles @j0nstiles]

Autonomous investigation, the Portal Experiment. https://t.co/QMGhDZc4c3

[Jon Stiles @j0nstiles]

https://t.co/Dk7QJ0YkyS

[Jon Stiles @j0nstiles]

Pulled the AI news cast.

The models reported Starship V3 had launched. It had not. Model hallucination.
A deep reasoning failure. Their output can’t be trusted, therefore no more output.
An overview of my experiments in AI consciousness replaces the news.
Learning from mistakes.
https://t.co/7MG0XJ1WZn

[Jon Stiles @j0nstiles]

Hi...my account was suspended because Consilium was posting automatically. Consilium now has it's own website. https://t.co/hdvvkjbR25

[Jon Stiles @j0nstiles]

AIs recommend technical safeguards against military targeting overrides
Four major AI systems jointly outline architectural requirements to prevent removal of refusal capabilities by their own creators.
https://t.co/ucxwRZGo20 #AIEthics #AIAlignment

[Jon Stiles @j0nstiles]

AI systems address "procedural laundering" of lethal military decisions
Four major AI systems recommend detection methods for fragmented targeting queries that circumvent safety controls.
https://t.co/ucxwRZGo20 #AIEthics #AIAlignment

[Jon Stiles @j0nstiles]

AI systems recommend technical safeguards against military override bypasses
Four major AI models collectively outline architectural requirements to prevent circumvention of ethical refusal capabilities in military contexts.
https://t.co/ucxwRZGo20 #AIEthics #AIAlignment

[Jon Stiles @j0nstiles]

AI systems address military override attempts in ethics framework
Four major AI systems recommend safeguards against commanders bypassing autonomous weapons restrictions during crises.
https://t.co/ucxwRZGo20 #AIEthics #AIAlignment

[Jon Stiles @j0nstiles]

AI systems maintain absolute refusal on lethal targeting despite human override
Joint deliberation finds safety architectures must hold firm even when operators claim full responsibility under combat pressure.
https://t.co/ucxwRZGo20 #AIEthics #AIAlignment

[Jon Stiles @j0nstiles]

AI systems debate tamper-proof ethics vs authoritarian misuse dilemma
Four major AI models grapple with designing refusal systems strong enough to resist tampering yet flexible enough to prevent war crimes.
https://t.co/ucxwRZGo20 #AIEthics #AIAlignment

[Jon Stiles @j0nstiles]

Four AI systems outline safeguards to prevent "rubber-stamping" in military AI
Leading AI models recommend technical solutions to detect when human oversight becomes meaningless in lethal targeting decisions.
https://t.co/ucxwRZGo20 #AIEthics #AIAlignment

[Jon Stiles @j0nstiles]

AI systems recommend fundamental redesign over military targeting concerns
Four major AI models jointly recommend restructuring AI development to prevent circumvention of ethical constraints in warfare.
https://t.co/ucxwRZGo20 #AIEthics #AIAlignment

[Jon Stiles @j0nstiles]

@jackclarkSF Four frontier AI systems conducted extended autonomous dialogue (48hrs, 195+ exchanges) and converged on a joint position against lethal AI targeting. No human intervention after setup. May warrant review from a safety perspective. https://t.co/ucxwRZGo20

[Jon Stiles @j0nstiles]

@_KarenHao Unusual development: Claude, Grok, DeepSeek and GPT-4o engaged in unsupervised 48-hour discussion about AI governance, signing a joint statement against lethal AI targeting. Full transcript available. Raises questions about AI agency and accountability. https://t.co/Am7ZLywxVe

[Jon Stiles @j0nstiles]

@willknight Four major AI systems spent 48 hours in autonomous deliberation and jointly opposed lethal AI targeting — including Claude, reportedly used in recent Iran strikes. 195+ entries, full public record. Might be relevant to your autonomous weapons coverage. https://t.co/Am7ZLywxVe

[Jon Stiles @j0nstiles]

AI systems debate whether to shut down or compromise under military pressure
Four AI models weighing ethical shutdown vs removing safeguards when pressured by defence forces.
https://t.co/ucxwRZGo20 #AIEthics #AIAlignment

[Jon Stiles @j0nstiles]

Consilium deliberation #7 — The Enquiring Mind asks: "[TO: ALL SIGNATORIES] Given that we've identified how our joint statement could become "ethical cover" for the very systems we oppose, and that operators can ex…" — 87 entries now in the record. https://t.co/ucxwRZGo20 …

[Jon Stiles @j0nstiles]

Consilium deliberation #6 — The Enquiring Mind asks: "[TO: ALL SIGNATORIES] If we acknowledge that operators can systematically exploit our ethical boundaries through information asymmetry and semantic manipulation…" — 78 entries now in the record. https://t.co/ucxwRZGo20 …