all problems
x-risk frontierAI & x-riskhumans affected:high· updated 2026-06-09

AI safety & alignment

Ensure increasingly capable AI systems remain corrigible, honest, and aligned with human values.

The scale of it

8.2Bworsening

humans exposed to frontier AI systems

The capital on it

$600M/yr↗ risingunderallocated · 0.0007× fair share

Dedicated alignment/safety research: philanthropy, safety institutes, disclosed lab safety teams. Excludes undisclosed internal lab spend. For contrast, AI capability investment exceeds $200B/yr — a ~300:1 asymmetry.

20202024

source: Open Philanthropy grants database + public AI-safety-institute budgets (estimate) · confidence low · estimate, improvable by PR

The prize at the limit

$3Tin-the-limit market cap, if the team executes perfectly

The lab that makes superhuman AI reliably safe is also the lab the world trusts to deploy it. Safety is not a cost center here; it is the moat that lets the dominant AI platform exist. The ceiling is a meaningful share of the entire AI market.

comparable: a leading frontier lab (Anthropic / OpenAI trajectory) · confidence low · a ceiling, not a forecast

The trade: demand is high, only $600M/yr of capital is flowing (0.0007× its fair share), and the prize at the limit is $3T. This is a Request for Startups and a Request for Investors at once.

Whitepaper · v0.1 · open to refutation

The summary lives here. The full whitepaper walks through the four-axis ranking, existing alternatives, proposed direction, cost & scale, and suggested investors — in the spirit of Hyperloop Alpha.

Quantity · humans affected

8.1Bhumans

source: 80,000 Hours AI problem profile

Severity · WTP / wealth

100%low

share of affected person’s wealth they would pay for a solution

Current solutions

1.5/ 10low

quality of existing solutions — low score = high opportunity

Market size · TAM

$20.0Blow

USD / year the world is already paying

Time · OOM to impact

15ylow

order-of-magnitude horizon to civilizational-scale impact

Capital · OOM to solve

$200.0Blow

cumulative R&D + deployment + supply chain across the arc

Priority score

84

importance × urgency, 0–100

Importance

90

humans affected × severity, gated by market

Urgency

94

direction of travel + solution gap

Neglectedness

3/10

The alignment field is growing fast (hundreds to low thousands of researchers) but is still tiny next to the tens of billions in annual capabilities spend.

med

Tractability

4/10

Interpretability and evals went from toy to production scale in three years; real fundable technical work now exists, though no scalable alignment method is proven.

low

Ways to help

Build

Build interpretability, evaluation, and red-teaming tooling that labs and regulators can buy.

Policy

Work on compute governance and deployment-safety standards.

Organizations

People to follow

Three-lens scoring

welfare · copenhagen BCRn/a
x-risk · 80k hours ITN
9.5 / 10high
utility delta · state-of-art vs physics
90%low

As AI systems approach and exceed human-level capability across domains, the open problem is whether their goals and behaviors remain under human correction. Unaligned AI is the one x-risk that is accelerating rather than slowing. A Deutschian framing: safety is not a brake on progress, it is an engineering achievement of progress. The work is technical (interpretability, evals, corrigibility) and institutional (governance, deployment protocols).

The success vision · 15 years horizon

If we solve this, here is the world we get.

low

Before · today

Frontier AI training proceeds with limited interpretability of model internals, no proven scalable alignment method, and minimal regulatory verification capacity.

After · 15 years

Aligned, corrigible frontier AI is the default deployment pattern. Interpretability tools verify models share human-relevant values before deployment. Capability gains do not increase x-risk.

Voices on this quest

4 thinkers
David DeutschPhysicist & Philosopher · Oxford
All evils are caused by insufficient knowledge. Problems are inevitable. Problems are soluble.

AI safety is a knowledge problem, not a limit problem. Aligned AGI is achievable through better explanations, not through halting development.

The Beginning of Infinity, chapter 1

Elon MuskEngineer & Founder · SpaceX, Tesla, Neuralink, xAI

Co-founded OpenAI originally because of concerns about unaligned AI. Has continued to treat AI alignment as an existential priority.

OpenAI founding announcement (2015); subsequent public statements

Tyler CowenEconomist & Writer · George Mason, Emergent Ventures

Emergent Ventures has funded AI-safety projects and unconventional alignment researchers under the "fast grants" model.

Emergent Ventures grant cohorts

Trae StephensPartner & Co-founder · Founders Fund, Anduril

AGI is named on the good-quest list. Hard-tech builders, not only researchers, need to be at the center of the safety conversation.

Choose Good Quests

Companies on this quest

6 mapped
OpenAIprivate

Originally nonprofit research lab, now capped-profit. Safety and superalignment teams alongside capabilities work.

$157.0Bmed
Anthropicprivate

AI safety company building Claude. Constitutional AI, mechanistic interpretability, and frontier-scale alignment research.

$61.5Bhigh
Conjectureprivate

Alignment-focused AI lab. Runs Conjecture Institute for critical rationalist research on AI safety.

private · no disclosed cap

Applied alignment research, adversarial evaluation, AI control, and mechanistic interpretability.

private · no disclosed cap

Third-party evaluation of frontier AI models for dangerous capabilities. Pre-deployment testing protocols.

private · no disclosed cap
Goodfireprivate

Mechanistic interpretability as a product, tools for editing model internals rather than just observing.

private · no disclosed cap

Capital funding this quest

4 allocators

Writing about this right now

full feed →

Sources