Strategy tags.

Lead with regulation, treaties, liability regimes

Solve technical alignment before capability thresholds close

Technology and markets solve risks faster than regulation creates them

AGI risk narratives overstated; real harms are mundane and current

Extinction/disempowerment risk overrides ordinary cost-benefit

Capability/risk evals gate deployment; evals are the load-bearing artefact

Release weights widely; transparency beats closed safety

also: open weights

Documented present harms outweigh speculative existential narratives

also: AI ethics

Build faster; delay costs more than capability

also: e/acc, effective accelerationism

Halt frontier training until alignment catches up

also: moratorium, stop-ai

Model welfare/moral status is a primary consideration

Arms-control-style treaty on frontier training or deployment

also: arms control

Mechanistic interpretability is necessary and sufficient to know models are safe

Break concentration via competition law

Build aligned superintelligence first, before adversaries

Control flops via export controls, licensing, reporting

Nation states must build their own AI for sovereignty

Decisions about AI must come through democratic processes

Responsible scaling policies; labs commit to capability-tied safety

Invest in AI-AI and AI-human cooperation capacities

Reject superintelligence as a goal entirely; narrow AI only

Build AI via open-ended self-generated curricula; safety must follow from the dynamics

also: autocurricula

Treat safety as adversarial security; assume systems break under attack

Preferentially develop protective technology over dangerous