Conditional commitments for AI safety coordination

New York, NY

spartacusapp.netlify.app

A platform for conditional commitments: pledges to act only when N peers agree, so lab employees, researchers, and policy coalitions can coordinate high-stakes collective action.

Project Details

What we're building:

One of the central challenges in AI safety is addressing coordination problems upstream of risks. Spartacus.app was and is designed as human coordination infrastructure, using assurance contracts to build a critical mass of participants to a designated action threshold, with optional temporary anonymity to protect users in adversarial conditions.

The problem it addresses:

AI safety and governance are hostage to several perverse incentive structures and game-theoretic problems, such as arms-race dynamics between labs and rival nations, competitive market dynamics that crowd out adequate support for guardrails, and proprietary R&D frameworks that inhibit information sharing as a public good.

The highest-leverage safety-oriented actions are deterred by first-mover and defection risk, not by a lack of latent support. Ultimately, this principle is scale-agnostic and can be applied even to the highest-stakes coordination dilemmas.

Functionality:

Two-sided conditional-commitment marketplace matching organizers and participants.

Public campaigns surface common knowledge about a community's latent preferences.

Transparent, strong commitment signals create preference cascades and bootstrap a critical mass of support for proposals.

Action-enforcement mechanisms (social accountability, financial deposits) strengthen commitment signals and filter out symbolic or expressive support.

Optional anonymity until threshold protects privacy and removes first-mover risk.

LLM-enhanced campaign builder calibrates campaign structure and expectations to optimize for likely success conditions.

Next steps:

Phase one of http://spartacus.app validated parts of the mechanism, namely the increased leverage resulting from the synchronized action of a large group, but the central question of whether the Assurance Contract incentive alone can induce coordination remains partially unresolved. Does conditional commitment convert private willingness into collective action without the need for a charismatic organizer or an external forcing function?

We have LOIs from AI safety-oriented organizations that are ready and willing to experiment, with no new strategies or engineering involved. Either signatories won't coordinate even with common knowledge of mutual gain (in which case the key thesis is flawed), or they will commit at a specific threshold and under reciprocal criteria when no one would act first alone (verifying the thesis).

Team:

Jordan Braunstein: Founder & CEO, full-time. Startup GTM operator who scaled venture-backed startups (Vivid Vision to $4.5M ARR, $10M+ at HaptX. Awarded ACX Grant 2024. Owns product, strategy, partnerships, pilot design, and case-study production.

Aster Langhi: Technical Lead. Over a decade of engineering experience across Google, Adobe, Afterpay (lead architect), and as a startup co-founder and CTO. Leads the rebuild, the AI-assisted matching layer, and security hardening.

CM: Technical Fellow, part-time. Engineering support.

LinkedIn profiles for core team

Jordan: https://www.linkedin.com/in/jordanbraunstein1

Aster: https://www.linkedin.com/in/seanlanghi/

Theory of Impact

Incentive structures govern human behavior in predictable ways, but can generate paradoxical or perverse effects when they optimize for one set of outcomes people want at the expense of other outcomes that those same people also want. Short-term interests can undermine long-term ones, and narrowly selfish preferences can backfire when pursuing them makes everyone's future self worse off than if they had been more cooperative from the outset.

Assurance contracts, or conditional commitments, help people overcome bad incentive structures and coordinate around preferences for long-term or large-scale benefits that would otherwise be blocked by cross-cutting pressures to pursue immediate gains.

We have a short window of opportunity (a few years at most) to solve key coordination and collective action problems that will determine humanity's path. The forks in the road that lead to wonderful or catastrophic outcomes are likely to be immediately in front of us.

There are several use cases among the AI researcher, operator, and policy-making communities that want to coordinate on higher-level collective goals but lack the coordination infrastructure to do so.

People

Jordan Braunstein

Team Member

Funding Details

Start Date: Feb 14, 2024
End Date: -
Expected Duration: Ongoing
Funding Raised to Date: -
Annual Budget: -
Monthly Burn Rate: -
Current Runway: -
Funding Goal: -
Funding Stage: Seeking additional support
Fiscal Sponsor: Manifold for Charity Inc. (Manifund)

Grants Received

Add grant

ACX Grants 2024

from ACX Grantsmanifund.org

$17,090

Funding Asks

grantmaking.ai Launch Round

Conditional commitments for AI safety related collective action

Applied

A platform for conditional commitments: pledge to act only when enough peers do, so lab employees, researchers, and coalitions can coordinate high-stakes collective action campaigns.

Minimum

$15,000

Ideal

$200,000

What this ask is

What we're building:

One of the central challenges in AI safety is addressing coordination problems upstream of risks. Spartacus.app was and is designed as human coordination infrastructure, using assurance contracts to build a critical mass of participants to a designated action threshold, with optional temporary anonymity to protect users in adversarial conditions.

The problem it addresses:

AI safety and governance are hostage to several perverse incentive structures and game-theoretic problems, such as arms-race dynamics between labs and rival nations, competitive market dynamics that crowd out adequate support for guardrails, and proprietary R&D frameworks that inhibit information sharing as a public good.

The highest-leverage safety-oriented actions are deterred by first-mover and defection risk, not by a lack of latent support. Ultimately, this principle is scale-agnostic and can be applied even to the highest-stakes coordination dilemmas.

Functionality:

Two sided conditional commitment "marketplace"; matching organizers and participants.
Public campaigns surface "common knowledge" about a community's latent preferences.
Transparent, strong commitment signals create preference cascades and bootstrap a "critical mass" of support for proposals.
-Various action enforcement mechanisms (social accountability, financial deposits) enhance commitment signal strength and filter out symbolic or expressive support.
-Optional anonymity until threshold protects privacy, and reduces "first mover" risk.
-LLM-enhanced campaign builder helps calibrate structure and expectations of campaign to optimize for likely success conditions.

Next steps:

Phase one of http://spartacus.app validated parts of the mechanism, namely the increased leverage resulting from the synchronized action of a large group, but the central question of whether the Assurance Contract incentive alone can induce coordination remains partially unresolved. Does conditional commitment convert private willingness into collective action without the need for a charismatic organizer or an external forcing function?

We have LOIs from AI safety-oriented organizations that are ready and willing to experiment, with no new strategies or engineering involved. Either signatories won't coordinate even with common knowledge of mutual gain (in which case the key thesis is flawed), or they will commit at a specific threshold and under reciprocal criteria when no one would act first alone (verifying the thesis).

Team:

Jordan Braunstein: Founder & CEO, full-time. Startup GTM operator who scaled venture-backed startups (Vivid Vision to $4.5M ARR, $10M+ at HaptX. Awarded ACX Grant 2024. Owns product, strategy, partnerships, pilot design, and case-study production.

Aster Langhi: Technical Lead. Over a decade of engineering experience across Google, Adobe, Afterpay (lead architect), and as a startup co-founder and CTO. Leads the rebuild, the AI-assisted matching layer, and security hardening.

Clarina M: Technical Fellow, part-time. Engineering support.

LinkedIn profiles for core team

Jordan: https://www.linkedin.com/in/jordanbraunstein1

Aster: https://www.linkedin.com/in/seanlanghi/

How the money will be spent

Minimum ($15,000): finish the refactor, ship one flagship case study.

This is the bare-minimum: it retains our Technical Lead, currently part-time at delivery-based compensation well below market, for roughly three months to complete the platform rebuild, and funds one extended, high-stakes campaign with our anchor LOI partner, Coalition for Good Futures (a cross-ideological AI-safety coalition statement), published as a case study with third-party attestation, in the mold of our SAG-AFTRA / AB 412 campaign. Covers infrastructure, hosting, LLM API costs, and Manifund's 5% fiscal sponsorship fee. Anything beyond this scope requires the next tier.

At this round's cap ($50,000): 1 to 2 case studies, ~6 months runway.

This matches the minimum / Speculation Grant scenario in our pending SFF S-Process application (budget attached as prior context): roughly six months of runway at subsistence compensation for the founder and Technical Lead, completion of the rebuild, and 1 to 2 published case studies from our LOI pipeline.

Ideal ($200,000): the full twelve-month plan.

This is the base scenario from a recent SFF application, covering June 2026 through May 2027:

Founder compensation (Jordan) — $40,000
Technical Lead (Aster) — $40,000
Student fellowships (1 + one TBD) — $5,000
Specialist contractors (security, design, legal) — $15,000
Marketing, events, travel — $15,000
Brand & UX redesign (one-time) — $10,000
Infrastructure (hosting, LLM API, third-party) — $15,000
Payment processing fees — $3,000
Compliance, accounting, legal — $5,000
Contingency — $10,000
Manifund fiscal sponsorship fee (5%) — $10,000
Reserve (pilot subsidies for partner orgs, opportunistic contractor sprints, contingency) — $32,000

At this level we target 5 to 10 published case studies.

Discussion

No comments yet. Be the first to share your thoughts.