Foundations7 min
What is p(doom)?
A clear guide to p(doom): what the shorthand means, why estimates vary, and why it is a personal judgment rather than a scientific measurement.
Read the guideP(Doomed) research archive · 01
Understand the theories behind AI catastrophe.
A source-led field guide to p(doom), p(hack), alignment, power-seeking, takeoff, and AI takeover scenarios. We separate technical ideas, open expert debates, and speculation—because they do not carry the same evidence.
06
independent guides
Built for curious readers, students, researchers, and players who want to inspect the assumptions beneath a scenario.
Reading paths
Choose your entry point
01
Start with the language
Foundations
Define p(doom), disambiguate p(hack), and understand what alignment researchers are trying to solve.
Begin here02
Trace the mechanism
Risk mechanisms
Examine why capability growth, power-seeking incentives, and takeoff speed change the shape of risk.
Begin here03
Compare the pathways
Scenario analysis
Separate malicious use, competitive races, organizational failure, and loss-of-control theories.
Begin hereFoundations6 min
What is p(hack)?
P(hack) is not a standard AI safety metric. Learn the two risks it can describe: AI-enabled hacking and advanced AI systems being compromised.
Read the guideFoundations8 min
The alignment problem
Understand the AI alignment problem: why specifying human intent is difficult, how proxies fail, and what researchers mean by aligned AI.
Read the guideRisk mechanisms7 min
Instrumental convergence
A balanced explanation of instrumental convergence, the orthogonality thesis, and why many different AI goals could produce similar power-seeking strategies.
Read the guideRisk mechanisms7 min
Fast vs slow takeoff
Compare fast and slow AI takeoff theories, why the speed of capability growth matters, and what each scenario implies for safety and governance.
Read the guideScenario analysis9 min
AI takeover theories
A grounded guide to AI takeover and doomsday scenarios, from malicious use and competitive races to loss of control and structural risks.
Read the guideFrom theory to systems
Test the assumptions in motion.
P(Doomed) is a speculative sandbox, not a prediction engine. Use it to explore how resources, factions, alignment failures, and human responses interact.