How to Raise $100M With Doomer Fanfiction
If you can’t explain your startup’s moat, don’t worry: you can always explain the apocalypse. Preferably in a pitch deck with a skull emoji you removed at the last second for “maturity.”
The reveal
The modern AI funding ecosystem has a recurring pattern: existential framing spikes attention, attention attracts capital and policy leverage, and suddenly everyone is "the only responsible adults in the room." Sometimes the adults even are responsible. The incentives still do what incentives do.
The steelman (yes, but also no)
Yes: some frontier risks are serious enough to justify real safeguards, independent evaluation, and governance. No: "civilization is at stake" is also a convenient story to tell when you want deference, regulatory carve‑outs, or a $1B round for something you can't fully describe yet.
The three incentive engines (the part nobody wants to call marketing)
Here's how the game works: Attention rewards memorable claims and strong emotions, which tends to produce "extinction risk" headlines with minimal mechanism. Capital rewards differentiation and inevitability, leading to pitches like "We're building the thing that ends history (safely)." Policy leverage rewards a crisis frame that justifies exceptionalism, resulting in arguments like "We must regulate… in a way only we can comply with."
The doom narrative is compute-efficient: it compresses complexity into a single number and lets everyone argue about the number instead of the mechanism. That's the real scaling law here.
Receipts (people actually saying the quiet part loudly)
Want proof? The CAIS "Statement on AI Risk" press release from May 2023 (https://safe.ai/work/press-release-ai-risk) is a canonical example of "extinction framing" becoming mainstream messaging. The Future of Life Institute's "Pause Giant AI Experiments" letter (https://futureoflife.org/wp-content/uploads/2023/05/FLI_Pause-Giant-AI-Experiments_An-Open-Letter.pdf) shows the pause campaign in action: real signatories, real influence, and a real tendency to jump from current systems to speculative endpoints. Reuters reported on Safe Superintelligence raising $1B in September 2024 (https://www.reuters.com/technology/artificial-intelligence/openai-co-founder-sutskevers-new-safety-focused-ai-startup-ssi-raises-1-billion-2024-09-04/), demonstrating how "safety" has become a brand and a fundraising narrative in its own right.
For contrast, OpenAI's Preparedness Framework v2 (https://cdn.openai.com/pdf/18a02b5d-6b67-4cec-ab64-68cdfbddebcd/preparedness-framework-v2.pdf) shows what "risk management" looks like when you have to operationalize it—sober, specific, measurable. Anthropic's Responsible Scaling Policy v3.0 (https://anthropic.com/responsible-scaling-policy/rsp-v3-0) demonstrates that voluntary commitments are real—but also reversible—because incentives don't stop at the press release.
How the grift happens (even when nobody is "a grifter")
You don't need villains. You need a competitive market, incomplete metrics, and PR that rewards certainty. The industry sells competence, regulators demand reassurance, and the public wants a story with a beginning, middle, and end. "We might all die" is a story. "We need better evals for bio‑capability uplift under tool access constraints" is a spreadsheet.
A doomer fanfic detector (questions that force mechanism)
Want to separate signal from noise? Ask: "What is the concrete harm pathway?" A serious answer includes steps, prerequisites, bottlenecks, and a credible actor model. Ask: "What is your disproof condition?" You should get a measurable failure mode for the claim itself. Ask: "What is your governance plan if competitors don't comply?" Look for incentive alignment, not just moral alignment. Ask: "What's your 'tripwire' capability threshold?" Demand observable tests tied to concrete mitigations.
If the answers are vibes and the slide deck has more flames than footnotes, you're not in "safety." You're in sales.
Disproof conditions (what would convince us the doom framing is mostly earned)
Here's what would actually justify the crisis tone: If we saw consistent, independently replicated eval evidence of capabilities crossing catastrophic thresholds, the warnings would become anchored in measurement. If we saw credible, enforced governance that binds labs even when it hurts revenue, the "we're responsible adults" claim would stop being PR. If we saw transparent incident reporting plus third‑party audits as a default, the field would start acting like it believes its own rhetoric.
Action line
When someone tells you to panic, ask them to cash it out into: a capability threshold, a likelihood estimate with reasons, and a mitigation plan that still works if their competitor is sprinting. If they can't do that, the probability is just astrology with a term sheet.
Anyway, back to reality—where the real grift is pretending incentives don't exist.
Related reading
Next up: /doom-astrology-chart, because "p(doom)" is not a substitute for mechanism. Then /regulatory-capture-for-dummies, because crisis narratives love becoming policy. Finally, /vibes-vs-evals, because measuring beats manifesting.