EF Security DAO
Grant Proposal

LeakDetectAI is an open-source Ethereum security framework designed to detect and measure hidden information leakage from transaction timing, gas usage, signing behavior, execution traces, and cross-chain interactions. It brings AI-driven side-channel analysis and adversarial inference techniques into Ethereum wallets, AI agents, and multi-chain applications, addressing a security gap that conventional exploit-focused tooling often fails to measure. With EF DAO Security funding, we will develop a production-ready CLI/SDK, reproducible public benchmarks and datasets, wallet and agent integrations, and standardized metrics for quantifying leakage and security risk. The project will deliver open-source research and practical tooling that helps Ethereum developers test whether their systems remain secure not only when executed, but also when observed by an adversary.

Ethereum security tools primarily focus on exploitable bugs, but provide limited ways to detect what a system unintentionally reveals through timing, gas usage, signing behavior, execution traces, and cross-chain activity. LeakDetectAI closes this gap by making hidden information leakage measurable through adversarial inference, machine learning, and reproducible security metrics.

Afterwards, developers can test wallets, AI agents, and multi-chain applications against leakage attacks before deployment, quantify the severity of observed leakage, compare mitigations, and reproduce results using shared datasets and benchmarks. This enables security testing based not only on whether a system executes correctly, but also on whether sensitive information can be inferred from its observable behavior.

Manu Gupta: Project Lead / Open-Source Security & P2P Maintainer. Manu is a long-time open-source developer and Technical Lead/Maintainer in the libp2p ecosystem, with 20+ years of open-source experience and work spanning decentralized networking, cryptography, Ethereum infrastructure, and security research. He leads LeakDetectAI’s Ethereum integration, open-source development, ecosystem coordination, and security tooling. Profile: https://github.com/seetadev

Dr. Pritha Gupta: Research Lead. Dr. Gupta brings research experience in AI, cryptography, side-channel analysis, and automated security testing, including work at Ruhr-Universität Bochum and Software Innovation Campus Paderborn. She leads the research methodology, adversarial inference experiments, ML/NAS pipelines, and security benchmarks underlying LeakDetectAI. Research: https://sec4ai-ai4sec.github.io/

Open-source and research relationships: LeakDetectAI builds on and extends publicly available research and codebases including AutoSCA, deep-learning side-channel analysis, and AutoML leakage-detection projects; these relationships are technical/open-source rather than ownership or employment relationships. The project has received support through Ethereum ecosystem security funding/QF participation and related open-source research grants; funding is used for open research and development, with code, benchmarks, and datasets intended to remain publicly accessible. Core research repositories include https://github.com/SEC4AI-AI4SEC/autosca-hw-wallet-tester and https://github.com/SEC4AI-AI4SEC/AutoMLQuantILDetect.

LeakDetectAI has a substantial head start in research, tooling, experiments, and ecosystem validation. The project builds on 3+ years of open-source security research and existing AI/side-channel analysis codebases, with ongoing work covering Ethereum, Zcash, Tornado Cash, wallets, cross-chain workflows, and P2P/libp2p network behavior. We have already developed automated approaches to detect, measure, reproduce, understand, and mitigate hidden leakage, and have begun building datasets and reproducible experiments.

The requested grant is for extending and hardening this existing foundation rather than building a security program from scratch. Existing research code, experiments, methodology, datasets, open-source infrastructure, and ecosystem relationships substantially reduce the engineering and research cost of producing the proposed Ethereum-native toolkit and public benchmarks. Grant funding will primarily support integration, validation, reproducibility, dataset development, documentation, security analysis, and open dissemination. This makes the proposed work materially less expensive than commissioning an equivalent program from scratch, while preserving the resulting tooling, benchmarks, and research as open public goods.

The grant will fund the transition of LeakDetectAI from research prototypes and experiments into reusable, reproducible Ethereum security infrastructure. Concrete deliverables include:

Funding primarily pays for research and engineering time to complete these integrations, run large-scale experiments, curate and document datasets, validate findings, maintain open-source repositories, and publish reproducible results.

The resulting code, benchmarks, documentation, and research outputs will remain openly accessible so that Ethereum developers and security researchers can independently reproduce and extend the work.

This grant is focused on open-source security research and reusable public infrastructure, not on building a proprietary security product or providing ongoing commercial security services. It deliberately does not include:

The grant also does not assume that every detected signal represents an exploitable vulnerability. The initial objective is to provide reproducible measurement, adversarial analysis, evidence, and mitigation guidance so that researchers and developers can independently assess the significance of observed leakage.

License: LeakDetectAI will remain open source, with core code, testing infrastructure, documentation, and reproducible benchmarks released under permissive open-source licenses. Research outputs and datasets will be published openly where licensing, privacy, security, and third-party rights permit.

Maintenance: We commit to maintaining the core repositories after the grant period, reviewing issues and contributions, fixing important bugs, and keeping the tooling usable with supported Ethereum environments. Grant-funded work will be developed in public repositories so that the community can inspect, reproduce, fork, and extend it independently.

Pinned targets: By the end of the grant period, we commit to delivering a usable LeakDetect CLI/SDK, documented leakage-detection workflows, reproducible Ethereum security benchmarks, public datasets where appropriate, standardized leakage metrics, initial wallet/AI-agent integrations, and published technical findings and mitigation guidance. Progress will be tracked through public repositories, releases, benchmarks, documentation, and research reports.

Exceptions: We are not requesting an exception to the open-source commitment. Where datasets contain sensitive information, third-party material, or information that could enable real-world exploitation, we may withhold or transform the affected data and publish reproducible methodologies, derived metrics, sanitized datasets, or equivalent test fixtures instead.