Project summary

Organizations and governments have strong incentives to automate open-source intelligence (OSINT), a task at which even non-specialized LLMs already perform well. This automated intelligence (AUTOINT) capability democratizes high-quality intelligence analysis, creating novel reasons to support aligned, interpretable AI [1] and new opportunities for macrostrategists to answer decision-relevant questions at scale [2].

This space is nearly empty; my paper [1] is the first on governing this technology.

I'm an Oxford Ellison Scholar (PNAS, ICLR publications; forthcoming NYT exclusive) on an unfunded gap year with ~7mo remaining. This grant covers ~2mo of living expenses to produce three concrete deliverables:

(a) Revise [1], currently in rebuttal at ICML (fixable critiques from reviewers), for publication at a top ML venue.

(b) Produce/circulate a memo scoping an AUTOINT research branch at a top AI policy org (in early talks).

(c) Moonshot: produce/circulate a proposal for an AI risk priority area at EIT [3] (well-resourced, influential org whose leadership has expressed openness to my input). >10% credence of success for <100hrs of work.

Goals

Terminal goals (in no particular order):

1. Reduce large-scale risks from AI deployment in high-stakes domains.

2. Increase strategic awareness [2] to improve macrostrategy.

Instrumental goals (within the next ~2 months) and theory of change:

1. Revise AUTOINT paper [1], aiming to publish it at ICML or a venue of similar quality. This publication would increase the popularity/prestige of advocacy for terminal goal #1 from a novel perspective.

2. Produce and circulate a memo at a top AI policy org advocating for an AUTOINT research branch. Such a branch would directly address important and neglected empirical governance questions, in line with terminal goal #2.

3. Produce and circulate a proposal for an AI risk priority area at EIT. Leadership has directly expressed openness to my input. Provides resources/credibility for terminal goal #1.

Track record [4]

• Publications at PNAS and ICLR; submission in review at ICML; news features at CBS, NYT (forthcoming), and more [5].

• Ellison Scholar, University of Oxford; Expert Collaborator, MIT AI Risk Initiative.

• Emergent Ventures grant recipient.

• NeurIPS referee.

• Advised World Bank/UN policy forecasts for 47 countries.

• Researched policy brief whose recommendation was implemented by UK and US governments (reported by NYT, BBC, White House).

On AUTOINT specifically: my paper [1] is, to my knowledge, the first academic work on governing automated OSINT. ICML reviewers’ critiques were fixable.

Failure modes

1. Paper fails to get published (e.g., ICML submission rejected w/o clear revision path)

   • Result:

     • Argument has less reach among top ML audiences (at least temporarily). Harms terminal goal #1.

   • Mitigations:

     • High-signal, truth-seeking communication with reviewers.

     • Resubmit elsewhere (e.g., relevant workshops at NeurIPS, ICLR, or AAAI).

2. Institutional backing is inaccessible (e.g., org scoping conversations suggest lack of interest):

   • Result:

     • Reduced expected resources for AUTOINT-powered strategic awareness. Harms terminal goal #2.

   • Mitigations:

     • Publish/share information value from scoping chats, and/or use it to motivate founding a new non-profit or startup.

3. EIT engagement doesn't convert

   • Result:

     • Meaningfully reduced expected resources for AI risk. Harms terminal goal #1.

   • Mitigations:

     • Prepared but private (DM/email [6] for details).

Past 12mo of funding

No other external funding to date.

Medium term, this agenda can absorb $100–300k+ for an organizational research branch or $1M+ for a startup.

[1] doi.org/10.48550/arXiv.2509.17087

[2] forethought.org/research/design-sketches-tools-for-strategic-awareness#automated-osint

[3] eit.org

[4] nicholaskruus.com/about

[5] nicholaskruus.com/papers

[6] nicholaskruus.com/contact