Is an AI takeover happening? What has actually been measured
What "AI takeover" means, what AI researchers said when surveyed, and what SIREN's hourly count of AI activity shows right now. Measurements, not predictions.
What does "AI takeover" mean?
In AI safety writing, a takeover means AI systems ending up in control of decisions that matter, whether by being handed that control or by taking it, in a way people cannot reverse. It is a scenario people argue about, not an event anyone has observed.
What do AI researchers say about it?
The largest survey so far asked 2,778 researchers who publish at top AI venues in October 2023. The median answer gave a 5% chance that advanced AI leads to outcomes as bad as human extinction; the mean was 16%. Between 38% and 51% of respondents put that chance at 10% or more. Those are opinions collected in a survey, and the authors say they are not a reliable guide to the truth. (Grace et al., Thousands of AI Authors on the Future of AI; AI Impacts)
What can actually be measured?
Activity. How many AI papers, model releases, data-centre deals, government rules and prediction-market moves happen, and whether that is more or less than usual. That is what SIREN counts every hour. A busy hour is not a takeover; it is a busy hour. Other people's risk estimates are collected on the p(doom) page, kept separate from the count.
What does SIREN read right now?
49.0 of 100 · level 4 (ROUTINE) · 2026-11-15 22:16:02 UTC
- Capability 29.3 of 100
- Compute & Capital 57.8 of 100
- Attention 37.9 of 100
- Governance 55.8 of 100
This is a count of AI activity against a fixed reference, not a probability and not a prediction. How is it computed? · Live gauge
Which newsroom headlines mention this now?
8 recent headlines about takeover, superintelligence, AGI or alignment, matched by keyword, so read before trusting the match.
- Alignment 'train/test splits' should be sacred 2026-11-15 18:22:03 UTC
- Former top national security advisor, Susan Rice, at a protest yesterday - "there is nothing more urgent than stopping Al from achieving super intelligence and causing human extinction" 2026-11-15 16:27:36 UTC
- In the early stages of Cadence, an open-source neural network architecture developed by Pragma Research, which sees to bring AGI to reality, here is a demo of a talking parrot who answers questions… 2026-11-15 16:12:08 UTC
- Motivations and incentives around AGI 2026-11-15 09:57:18 UTC
- Visualizing the Difficulty of AI Alignment 2026-11-15 06:27:15 UTC
- Visualizing the Difficulty of AI Alignment 2026-11-15 05:58:28 UTC
- One Router Grounded America? Seriously? Are you sure it wasn’t Rogue AI or a Hack? 2026-11-14 20:54:09 UTC
- AGI (Artificial General Intelligence) Lockdown and the End of the World: Why We Are Waiting for the Apocalypse Again 2026-11-14 16:38:25 UTC
Quick answers
Is an AI takeover happening?
No one has observed one. It is a scenario debated in AI safety research. What can be measured is AI activity, which SIREN counts every hour.
What do AI researchers think the risk is?
In a 2023 survey of 2,778 AI researchers, the median respondent gave a 5% chance of outcomes as bad as human extinction from advanced AI; the mean was 16%. These are opinions, not measurements.
Sourced facts checked 2026-11-14. SIREN measures how much is happening, not how bad it is. Levels run from 5 (quietest) to 1 (loudest) and describe observable activity tempo against this index’s own record. They are not a probability of harm.