You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$35,000 left to reach the funding goal.
Frontier Risk Institute is seeking $35,000 to conduct independent, open-access research into the safety and accountability challenges created by increasingly autonomous AI systems.
As AI agents gain the ability to access sensitive information, execute transactions, interact with external services, and make decisions on behalf of people and organizations, existing identity and authorization systems face new challenges.
Our project will investigate how autonomous agents establish identity, receive delegated authority, operate within defined permissions, and remain accountable for their actions.
We will publish an open Agent Accountability Framework, a taxonomy of emerging failure modes, and practical recommendations for developers, researchers, and organizations working to deploy AI agents safely.
The research will be freely accessible and developed independently of commercial vendors.
Our primary goal is to make emerging safety risks involving autonomous AI agents easier to identify, evaluate, and mitigate.
We will focus on four research questions:
What identity, authorization, and accountability failures become possible when AI agents operate autonomously?
How do existing identity and access management systems fall short when agents delegate tasks, interact with other agents, or operate across services?
What safeguards can reduce unauthorized actions, excessive permissions, privilege escalation, and failures of human oversight?
How can these findings become a practical, open framework for evaluating AI agent safety?
The project will follow a four-month research and publication process.
Month 1: Research and risk identification
Review existing technical literature, agent architectures, identity standards, authorization models, and documented failure scenarios. Develop an initial taxonomy of agent-related risks.
Month 2: Framework development
Analyze safeguards and develop the first version of the Agent Accountability Framework, covering identity verification, delegated permissions, monitoring, auditability, and human oversight.
Month 3: Expert review and validation
Seek feedback from relevant AI safety, cybersecurity, and identity specialists. Refine the framework, document its limitations, and develop representative examples showing how it can be applied.
Month 4: Publication and distribution
Publish the research report, risk taxonomy, framework, and implementation guidance. Make the outputs freely accessible and invite public feedback.
We will evaluate the project against specific deliverables and evidence of usefulness.
Research deliverables:
One publicly available research report examining identity, authorization, and accountability risks in autonomous AI systems.
One structured taxonomy of agent-related failure modes, supported by documented examples and references.
One open Agent Accountability Framework with practical evaluation criteria.
One implementation guide showing how developers and organizations can apply the framework.
A public repository or research library containing the findings and supporting documentation.
Quality and usefulness:
We aim to obtain substantive feedback from at least three relevant technical experts and incorporate that feedback into the final publications.
We will also track whether researchers, developers, or organizations reference, discuss, test, or apply the framework.
Success will not be measured primarily by website traffic or social media engagement. The most important outcome is producing credible, actionable research that improves understanding of emerging AI safety risks.
We are seeking $35,000 to support a four-month research initiative.
The proposed budget includes $20,000 for principal researcher compensation, $8,000 for specialist research assistance and independent technical review, $3,000 for research tools, data, and software, $2,000 for publication and public research infrastructure, and $2,000 for administration and compliance.
Principal researcher compensation will support research design, literature review, risk analysis, framework development, expert engagement, and publication. Any founder compensation will be subject to independent board approval and applicable nonprofit requirements.
The minimum viable funding amount is $15,000. At this level, we would conduct a narrower three-month project focused on agent identity and delegated authorization, producing an initial risk taxonomy, preliminary accountability framework, and open-access research report.
Additional funding would expand the research scope, strengthen external technical review, and support more detailed implementation guidance.
Brian Matthew Davis, Founder and Executive Director, Frontier Risk Institute
LinkedIn: https://www.linkedin.com/in/briandavis21/
Research platform: https://thehouseoffraud.com
Brian has experience leading fraud prevention and risk programs across fintech, e-commerce, and digital marketplaces, including serving as Head of Fraud at Dodgeball before its acquisition by Spreedly.
He founded The House of Fraud, a research and publishing initiative examining emerging developments in fraud, digital identity, payments, and artificial intelligence.
He currently serves as VP of Marketing and Community at Safeguard, working with senior practitioners across AI, fraud, cybersecurity, identity, risk, and compliance.
His experience combines operational knowledge of digital trust and risk systems with research, analysis, and communication of emerging technology developments.
Frontier Risk Institute is a newly established Massachusetts nonprofit. This will be one of its first formal research initiatives.
The project will initially be founder-led. Grant funding will support external technical reviewers and specialist contributors, who have not yet been formally engaged.
We do not claim an established academic AI safety research track record. Our advantage is practical experience with the identity, authorization, and risk systems that autonomous AI agents are beginning to interact with, combined with a commitment to independent, publicly accessible research.
The primary risk is producing research that is conceptually useful but insufficiently technical or actionable for AI developers and safety researchers.
A second risk is attempting to cover too many aspects of autonomous AI safety within a limited budget and timeline.
A third risk is insufficient engagement from technical experts, resulting in a framework that lacks adequate external scrutiny.
Finally, as a founder-led project, execution could be affected by competing responsibilities or insufficient research capacity.
We will mitigate these risks by narrowing the initial scope to identity, authorization, and accountability; establishing specific monthly milestones; seeking technical feedback before publication; and documenting the limitations of our findings.
If the project fails, the likely outcome is that we publish a limited or insufficiently validated research framework that does not meaningfully influence technical practices.
We would still make completed research publicly accessible, explain its limitations, and document what we learned.
If funding or capacity proves insufficient, we will prioritize a smaller, technically credible research output rather than attempt to deliver an overly ambitious framework.
$0.
Frontier Risk Institute is a newly established nonprofit and has not yet received grant funding or charitable donations.
We have submitted applications to several AI safety and public-interest research funders, but no funding has been awarded or received.
This Manifund proposal seeks initial funding for a defined research initiative. Any funding received from other sources for overlapping activities will be disclosed, and project budgets will be adjusted to avoid double funding.
There are no bids on this project.