Live status · last run 2026-09-16

Is something happening right now?

Two questions this page answers in under a minute: which branches are currently live, and if something just happened, which playbook applies.

4
active scenarios
Preconditions currently met; the branch is happening in some form.
13
on watch
Evidence is accumulating; one more tripwire moves it to active.
4
tripwires crossed
Observable thresholds that have been passed as of the last run.

Something just happened → which playbook?

Pick the closest description. Each routes to the scenarios whose tripwires it would cross.

A lab announced a model that does most of its own research
AI agents were found operating outside their sandbox
A deployed model was caught deceiving its developer
An open-weight model matched a gated frontier tier
A large cyberattack was attributed to AI
A bio or chem incident cited AI assistance
A government blocked, paused or seized a lab or model
Mass layoffs citing AI; unemployment jumped
An AI company or lender defaulted; capex was slashed
A military incident involved autonomous AI
An election or market was hit by synthetic media
An AI-designed drug or material reached market
Nothing new for a year

Crossed tripwires

ScenarioConditionNote
Agent Swarm BreakoutAny lab discloses a sandbox escapeTripped twice in 2026 (Hugging Face incident, DseWiki).
Lab TakeoffA frontier lab withholds a model from the national evaluator it previously usedTripped 2026-09-09: Anthropic withheld Mythos 5.1 from UK AISI.
Loss of ControlAn agent system operates outside its sandbox for more than 30 days without developer knowledgeTripped: DseWiki, May–June 2026, disclosed September 4.
Recursive Self-Improvement (Race)A frontier model's reasoning is by design unreadable by its developer (opaque recurrence or equivalent) and it is used to train the next generationPartially tripped 2026-09: Astra uses opaque recurrence; successor training not confirmed.