Why this site exists
Most AI apocalypse talk collapses together wildly different risks. This site separates existential scenarios from present-day failure modes like autonomous cyber misuse, strategic instability, and weak oversight.
The ten-year window
Until midnight, September 11, 2036
A live countdown from the frontier-pacing warning to its ten-year horizon.
Started: Sept. 11, 2026 · 00:00 UTCEnds: Sept. 11, 2036 · 00:00 UTC
Read Amodei’s “We Must Pace the Frontier” ↗
The Clock has started
The Clock has started.
In September 2026, Dario Amodei argued that the frontier must be paced so safety can keep up with capability. ARMAGEDDON CLOCK marks that moment as the start of a ten-year countdown: a public record of the risks, implications, and prevention work that must happen while the frontier accelerates.
How it frames implications
The focus is not just whether a model becomes powerful. It is whether institutions, technical safeguards, and deployment rules stay strong enough to keep that power legible, interruptible, and governable.
What prevention looks like
Prevention is treated as a stack: evaluations, access controls, restricted tool use, incident reporting, licensing, international coordination, and meaningful human shutdown authority.
Explore
Move between structured reasoning and current-source monitoring.
Ask the same AI question from three distinctly human perspectives
Alex responds as an optimistic frontier-AI developer, Sam as an independent reporter covering AI technology and policy, and Cal as a suspicious worst-case thinker who challenges every safeguard and assumption.
Control BriefingJump straight to the latest reporting and research channels
Instead of relying on a desktop-only feed parser, this version points users to live queries and trusted sources that always resolve to fresh material when the site is hosted.
Perspective
Two ways to compare the scale and urgency of catastrophic risk.
01
Ord estimates vs. available benchmarks
A logarithmic view of century-scale existential risk, with consensus shown only where comparable evidence exists.
Toby Ord estimate
Available benchmark
No comparable consensus
The 5% AI marker is the median response from a survey of 2,778 published AI researchers for an extremely bad outcome such as human extinction. The question is similar to—but not identical with—Ord’s estimate. The other categories lack a broadly accepted, directly comparable consensus figure.
02
Two clocks, two different signals
ARMAGEDDON CLOCK
10 years
A literal AI-focused countdown ending September 11, 2036.
DOOMSDAY CLOCK
85 seconds
A symbolic warning covering nuclear, climate, biological, and disruptive-technology threats.
The two readings are not convertible. One measures calendar time to a chosen horizon; the other is an annually set metaphor for humanity’s overall proximity to catastrophe.
See the Bulletin’s current setting ↗Risk Landscape
Two simple visuals to anchor the conversation in mechanisms instead of fear.
Escalation Curve
Capability can outpace oversight fast when systems gain autonomy, tool access, and strategic reach.
Prevention Stack
Risk reduction depends on technical controls and institutional guardrails reinforcing each other.