We track what 40 AI companies say in public about catastrophic risk, and score each statement on one five-step scale from Doom to Dismissive. Each company sits in the layer of the AI stack where its revenue comes from. A search runs every morning, the most confident findings publish straight away, and every placement links to its source.
Read the analysis behind the tracker: Doom is a frontier-lab position.
Ask us about making the most of AI in your business. Email hello@softwired.com.au or book a call.
The four companies that sell closed models at the frontier: OpenAI, Anthropic, Google DeepMind and xAI. Their leaders state a material probability of extinction or loss of control, or back slowing frontier development. At launch on 8 October 2026, none of the other 36 companies we track did. The summary at the top of this page gives today's count.
It is the mean position of every company we can place, on a scale from −2 (dismisses catastrophic risk) to +2 (doom), with each company counted equally. Because only the four frontier labs sit at Doom, the score stays below neutral unless the rest of the industry moves toward them.
Each public statement is Doom, Control, Operational or Dismissive. A company takes the step with the most weight behind it across its statements from the past two years, and a company with too little recent evidence shows as Silent. A statement from the CEO, a founder, the chair or the company itself counts in full, one from another senior executive counts half, and one from other staff a quarter. Every statement's weight halves every 90 days, so one CEO remark keeps a company placed for about ten months, and one from other staff for about four. If a company explicitly withdraws a position it held, its earlier statements stop counting once we confirm the reversal. The method follows InfluenceMap's scoring of corporate climate-policy positions.
A public, attributable remark by a named executive, founder or the company itself about catastrophic, extinction-level or loss-of-control AI risk, or about pacing frontier development. Remarks about jobs, privacy, security or liability count as Operational. We take statements from primary sources or tier-one outlets only, and every placement links to its source.
Every day. A search runs at 6:30 am Brisbane time and a classifier scores each statement it finds. Statements we classify with 0.8 confidence or more publish straight away. Less confident ones wait for review, but if several point the same way for one company they count towards its position before review. Their text isn't shown until someone approves it. History runs from October 2023.
Yes. The data is published under CC BY 4.0, so you can reuse it for any purpose with credit to the Softwired AI risk sentiment tracker and the date you accessed it. Download the current placements or the daily history as CSV.
Written by Softwired research, with James Hallam as research lead. This is version 1.1, last changed in October 2026. You can reuse the data under CC BY 4.0 with credit.
Softwired research (2026). AI risk sentiment tracker, version 1.1. Softwired. https://softwired.com.au/ai-sentiment. Licensed under CC BY 4.0. Accessed [day month year].
@misc{softwired_ai_sentiment_2026,
author = {{Softwired research}},
title = {AI risk sentiment tracker},
year = {2026},
version = {1.1},
publisher = {Softwired},
url = {https://softwired.com.au/ai-sentiment},
note = {Data under CC BY 4.0. Accessed YYYY-MM-DD}
}1.1, October 2026. A company's position now comes from the weighted balance of its statements over the past two years. A statement's weight depends on the speaker's seniority and halves every 90 days. Silent now means a company's statements carry too little weight to place it; before, it meant we had found no statement. We added a second scale for how good or bad each company expects the AI future to be, and the trend chart now shows daily points.
1.0, 8 October 2026. The tracker launched with 40 companies, the five-step scale and history back to 1 October 2023. Each company's position came from its latest approved statement, and the trend chart showed weekly points.
1.1, October 2026. A company's position now comes from the weighted balance of its statements over the past two years. A statement's weight depends on the speaker's seniority and halves every 90 days. We added a second scale for how good or bad each company expects the AI future to be, and the trend chart now shows daily points.
1.0, 8 October 2026. The tracker launched with 40 companies, the five-step scale and history back to 1 October 2023.