Wolfe Research A Brief History of the End of the World

The Daily Froth: A Brief History of the End of the World September 15, 2026

History offers some reassurance. Humanity has repeatedly faced predictions of catastrophe from new technologies, and the worst outcomes generally haven't materialized. But the reasons vary meaningfully: sometimes the fear was wrong, sometimes intervention helped prevent the worst-case outcome, and sometimes the technology did create a dangerous new capability. The difficult question today with AI is that we don't yet know which historical precedent applies, but we strongly suspect 10% odds of doomsday is quite overstated. We discussed the broad AI risks and possible responses over the weekend (link), and Wolfe’s Tobin Marcus followed with a detailed look at the policy outlook (link). With limited appetite from the White House for broad regulation (including comments from the President yesterday) and policy moving slowly regardless, much of the responsibility for managing frontier risks remains in the hands of the private sector. OpenAI and Anthropic have publicly committed to advancing AI with safety at the forefront. But one of the biggest challenges is how to pace frontier development without losing the technological edge to China, which has a very different set of objectives.

We took a brief look at catastrophic fears of the past, and found three distinct outcomes: fears that proved unfounded (Large Hadron Collider), risks that were real but contained by intervention (recombinant DNA), and technologies that genuinely created dangerous new capabilities that humanity still manages today (nuclear weapons). Y2K is more ambiguous -- the disruption was largely nonexistent, perhaps because of the expensive remediation effort or because there was no meaningful risk to society. If only Americans held onto their stockpiles of food and toilet paper for the pandemic twenty years later.

Where does that leave investors? We discussed three main risks in our weekend note with increasing severity: 1) Humans use AI in harmful ways, 2) AI does something we didn't ask it to do, and 3) humans lose control, and stand in the way.

Stephanie Roth Felipe Montoya Fiona Van Der Steur View Stephanie’s Research

This report is limited solely for Wolfe Research’s clients. Please see analyst certifications and important disclosures beginning on page 3. Please help us protect your advantage... DO NOT FORWARD Page 1 of 4

Cybersecurity firms stand to benefit from Risks #1 and #2, while AI safety and testing firms (although most remain private) and cloud hyperscalers should benefit from greater spending around Risk #2. Hard assets, resilient physical infrastructure and potentially Treasuries offer some hedge against Risk #3, although if we actually reach the point where humans lose control, investment returns probably aren't the biggest concern.

Please help us protect your advantage... DO NOT FORWARD Page 2 of 4