// Ethereum security · research & development

LeakDetectAI

Detecting and Eliminating Side-Channel Vulnerabilities in Ethereum Wallets and AI Agents

From reproducible information-leakage experiments to maintainer-accepted fixes, merged patches, and regression tests: a repeatable pathway to stronger Ethereum security.

Project lead
Manu Sheel Gupta
Research lead
Dr. Pritha Gupta
Recipient
SEETA
Expected duration
Approximately 15 months

Funding at a glance

Funding goal
$120,000
Existing pledges
$68,000
Remaining requirement
$52,000

Six planned milestones, each allocated $20,000. Delivery targets and prospective adoption are not completed outcomes or guaranteed partnerships.

Project Overview

LeakDetectAI is an open-source Ethereum security initiative focused on identifying, reproducing and eliminating side-channel vulnerabilities in Ethereum wallets, hardware wallets and AI-agent frameworks.

As Ethereum applications increasingly rely on cryptographic keys, transaction signing, automated authorization and autonomous AI agents, security testing must extend beyond conventional functional testing and smart-contract auditing. Observable execution behaviour, including timing and other information leakage, can expose sensitive information and create attack opportunities that are not adequately covered by traditional security testing.

LeakDetectAI addresses this gap through automated side-channel analysis, machine-learning-based detection, reproducible security experiments and direct remediation engineering.

The project builds on existing research, open-source tooling and experimental infrastructure. Its focus is to extend these capabilities into real Ethereum applications and development workflows, working directly with ecosystem developers and maintainers to identify vulnerabilities and implement verifiable fixes.

The primary objective is measurable security improvement across Ethereum: identified vulnerabilities, maintainer-accepted fixes, merged patches and regression tests.

The Security Problem

Ethereum wallets, hardware wallets and AI-agent frameworks increasingly manage sensitive cryptographic operations and authorization workflows.

However, developers lack accessible and reproducible methods to identify and remediate side-channel vulnerabilities in these environments.

Conventional security testing generally prioritises:

  • Smart-contract vulnerabilities and known attack patterns.
  • Functional correctness and application-level security.
  • Cryptographic implementation correctness.
  • Conventional penetration testing and vulnerability assessment.

These approaches can leave weaknesses arising from observable execution behaviour insufficiently tested.

Side-channel vulnerabilities can potentially expose sensitive information through timing and other observable characteristics of cryptographic operations and application workflows.

This creates several security concerns:

  • Wallet security: Exposure of information associated with key management and cryptographic operations.
  • Transaction signing: Potential leakage during signing and authorization workflows.
  • Hardware-wallet security: Insufficient testing of real-world device and signing interactions.
  • AI-agent security: Information leakage during agent execution, authentication and transaction authorization.
  • Ethereum Layer 2 ecosystems: Security weaknesses across interconnected wallet, protocol and agent environments.

LeakDetectAI provides a practical framework to identify these weaknesses, reproduce them and help developers implement effective mitigations.

Technical Approach

LeakDetectAI combines existing side-channel analysis research with automated machine-learning techniques and Ethereum-specific security testing.

The technical approach consists of five complementary components.

A. Automated Side-Channel Analysis

Extend existing automated side-channel analysis capabilities to Ethereum wallet and AI-agent workflows.

B. Machine-Learning-Based Detection

Apply deep-learning architectures, automated machine learning (AutoML) and neural architecture search (NAS) to detect and analyse information leakage.

C. Quantitative Information-Leakage Detection

Use information-theoretic methods to quantify potential leakage and provide reproducible evidence for security assessments.

D. Ethereum-Specific Testing

Apply the detection capabilities to desktop wallets, mobile wallets, hardware wallets, transaction signing, key-management operations and AI-agent execution environments.

E. Vulnerability Remediation and Verification

Collaborate with participating developers and maintainers to:

  • Reproduce confirmed vulnerabilities.
  • Prepare actionable technical reports.
  • Develop and implement fixes.
  • Submit or support upstream patches.
  • Build regression tests.
  • Verify that mitigations address the identified weaknesses.

The objective is to integrate repeatable security testing into existing development processes rather than provide a one-time security assessment.

Existing Research and Open-Source Foundation

LeakDetectAI does not propose building foundational side-channel analysis capabilities from scratch.

The project builds on existing research and open-source implementations, including:

AutoSCA

Automated side-channel analysis tooling developed in the context of Achelos, demonstrating automated analysis and machine-learning-based detection of cryptographic side-channel vulnerabilities, including Bleichenbacher-style TLS vulnerabilities.

Repository: https://github.com/achelos/AutoSCA ↗

Hardware-Wallet Security Testing

https://github.com/SEC4AI-AI4SEC/autosca-hw-wallet-tester ↗

An existing research and development foundation for automated side-channel security testing of hardware wallets.

Deep-Learning Side-Channel Analysis

https://github.com/SEC4AI-AI4SEC/deep-learning-sca ↗

A research and reproducibility codebase for automated side-channel analysis.

Quantitative Information-Leakage Detection

https://github.com/SEC4AI-AI4SEC/AutoMLQuantILDetect ↗

An existing package applying AutoML and information-theoretic techniques to detect and quantify information leakage.

The underlying research direction is also documented in the ACM AISec paper:

Automated Detection of Side Channels in Cryptographic Protocols: DROWN the ROBOTs!

LeakDetectAI extends this research foundation into Ethereum wallets, transaction signing, hardware-wallet workflows and AI-agent applications.

Ethereum Ecosystem Adoption

The project focuses on integrating LeakDetectAI into actual Ethereum security and development workflows.

Six adoption areas are prioritised.

5.1 Wallet and Hardware-Wallet Security

Test desktop, mobile and hardware wallets across representative Ethereum key-management, signing and authorization workflows.

Prospective adoption targets include:

  • MetaMask
  • Rabby Wallet
  • Trust Wallet
  • Safe

5.2 Existing Ethereum Field Deployments

Apply LeakDetectAI in existing Ethereum field-testing environments, including Ledger hardware-wallet workflows and the Diagnosity health-AI deployment context.

5.3 Ethereum and Layer 2 Ecosystems

Extend security testing across Ethereum Layer 1 and Layer 2 environments, including Optimism, Arbitrum and Base.

5.4 Ethereum Agent and Protocol Security

Evaluate side-channel risks in agent execution, transaction authorization and protocol-integrated workflows.

Relevant ecosystem contexts include Agent0 (ERC-8004), Aave and Morpho.

5.5 AI-Agent Framework Security

Engage agent-framework developers to test cryptographic operations, transaction execution and authorization workflows.

Prospective targets include:

  • Coinbase AgentKit
  • ElizaOS

5.6 Ecosystem-Wide Remediation

Consolidate findings into actionable remediation efforts, upstream patches, regression tests and publicly verifiable security improvements.

These organisations and projects represent prospective adoption targets and existing technical or ecosystem contexts, not guaranteed partnerships or adoption commitments.

Team

Manu Sheel Gupta — Project Lead

Manu brings over 20 years of open-source engineering experience spanning decentralized networking, cryptographic infrastructure, distributed systems and security engineering.

He is a Technical Lead and Maintainer in the libp2p ecosystem, with contributions to IPFS, IPLD and Multiformats across Go, Rust, JavaScript and Python implementations.

His responsibilities include:

  • Project delivery and technical coordination.
  • Ethereum and wallet ecosystem integration.
  • Open-source engineering and infrastructure.
  • Engagement with wallet and agent-framework maintainers.
  • Vulnerability remediation and ecosystem adoption.

GitHub: https://github.com/seetadev ↗

Dr. Pritha Gupta — Research Lead

Dr. Pritha Gupta brings research experience in artificial intelligence, cryptography, side-channel analysis and automated security testing, including work associated with Ruhr-Universität Bochum and Software Innovation Campus Paderborn.

She leads:

  • Technical research methodology.
  • Adversarial inference experiments.
  • Machine-learning and neural architecture search pipelines.
  • Side-channel detection methodology.
  • Security benchmarks and reproducibility.
  • Translation of research capabilities into practical testing workflows.

Research Profile: https://sec4ai-ai4sec.github.io/ ↗

Ecosystem Relationships and Deployment Context

LeakDetectAI builds on existing technical relationships and deployment contexts across Ethereum security, decentralized infrastructure, AI and open-source ecosystems.

Relevant relationships and technical contexts include:

  • Achelos / AutoSCA: Automated side-channel analysis tooling and research foundation.
  • Park Pro Token: Field-testing context involving Ledger hardware-wallet workflows and Ethereum activity.
  • Diagnosity: Health-AI observability pilot and field-deployment context.
  • Agent0 (ERC-8004): Trustless agent discovery, reputation and validation alignment.
  • Aave and Morpho: Existing enterprise and Health-AI observability engagement contexts.
  • Thales Security: Security research and advisory relationship.
  • libp2p, IPFS, IPLD and Multiformats: Decentralized networking, content-addressed storage and interoperable infrastructure.
  • SAP BTP and Joule: Enterprise AI-agent observability engagement.
  • Samagra: Public-sector deployment experience.
  • Public AI: Open and verifiable AI infrastructure alignment.

These relationships provide potential pathways for testing, validation and adoption. They should not be interpreted as formal commitments from every organisation to adopt LeakDetectAI.

Milestones and Deliverables

Planned adoption milestones · January 2027 — March 2028 · $20,000 per milestone

  1. January 2027

    $20,000 · Milestone 1

    Desktop, Mobile and Hardware-Wallet Security Testing

    • Deploy LeakDetectAI in participating Ethereum Ledger hardware-wallet testing workflows.
    • Validate desktop, mobile and hardware-wallet environments.
    • Document reproducible security weaknesses.
    • Support at least one maintainer-accepted fix or mitigation.
    • Publish a technical deployment report and practical security recommendations.
  2. March 2027

    $20,000 · Milestone 2

    Existing Ethereum Field Deployments

    • Integrate LeakDetectAI into the Diagnosity Ethereum health-AI application workflow.
    • Demonstrate actual use across at least three wallet implementations or development teams.
    • Deliver reproducible security findings.
    • Support remediation and regression testing.
    • Publish verifiable adoption evidence.
  3. May 2027

    $20,000 · Milestone 3

    Ethereum Agent and Protocol Ecosystems

    • Integrate into at least two participating Ethereum agent or protocol environments.
    • Prioritise Agent0, Base, Aave and Morpho.
    • Test agent execution, transaction and signing workflows.
    • Engage at least two independent development or security teams.
    • Secure at least one maintainer-accepted fix or mitigation.
  4. July 2027

    $20,000 · Milestone 4

    Ethereum Wallet Developer Adoption

    • Engage MetaMask, Rabby, Safe and Trust Wallet as prospective adoption targets.
    • Integrate into at least two wallet or agent-framework development workflows.
    • Deliver maintainer-reviewed vulnerability reports.
    • Secure at least one merged patch or verifiable downstream fix.
    • Provide regression tests.
  5. September 2027

    $20,000 · Milestone 5

    Ethereum Agent-Framework Adoption

    • Engage Coinbase AgentKit and ElizaOS.
    • Integrate into at least two agent-framework development workflows.
    • Identify and reproduce side-channel security weaknesses.
    • Deliver remediation recommendations.
    • Secure at least one merged patch or verifiable downstream fix.
    • Publish regression tests and share findings through ecosystem workshops.
  6. March 2028

    $20,000 · Milestone 6

    Ecosystem-Wide Adoption and Verified Remediation

    • Achieve verified adoption in at least three named Ethereum ecosystem projects.
    • Demonstrate security remediation in at least three participating codebases.
    • Include at least one Ethereum wallet and one AI-agent framework.
    • Provide verifiable evidence through merged pull requests, commits, releases or equivalent downstream changes.
    • Publish an adoption report connecting identified vulnerabilities, implemented fixes and verification evidence.

Measurable Impact

LeakDetectAI measures success through implemented security improvements rather than downloads, user counts or demonstration activity.

The principal targets are:

  • Adoption by at least three Ethereum wallet or agent-framework projects.
  • At least three implemented security improvements across participating codebases.
  • At least one wallet security remediation.
  • At least one AI-agent framework security remediation.
  • Reproducible vulnerability reports and technical evidence.
  • Regression tests or equivalent validation accompanying accepted remediations.
  • Public documentation and reusable open-source testing infrastructure.

Each counted remediation must be linked to a reproducible weakness identified through LeakDetectAI and supported by verifiable evidence of implementation.

Sensitive findings will be managed through responsible-disclosure processes.

Funding Request and Allocation

Total project budget: $120,000

Existing ecosystem pledges: $68,000.

Remaining funding requirement: $52,000.

The project has received community support through the Giveth / Ethereum Security Quadratic Funding round, including community contributions and matching-pool support.

The funding request supports the next stage of development and adoption:

  • Ethereum wallet and hardware-wallet testing.
  • Integration into existing field deployments.
  • Ethereum Layer 1 and Layer 2 ecosystem engagement.
  • AI-agent framework security integrations.
  • Vulnerability reproduction and remediation engineering.
  • Regression testing and independent verification.
  • Open-source documentation and ecosystem adoption.

The six milestones are each allocated $20,000, with funding tied to verifiable delivery and acceptance.

Open-Source and Public-Goods Commitment

LeakDetectAI is committed to making grant-funded, non-sensitive software and research outputs publicly available.

The project commits to:

  • Releasing grant-funded detection tooling, benchmarks, test cases and documentation under an OSI-approved open-source licence, preferably Apache-2.0.
  • Maintaining public GitHub repositories and development history.
  • Publishing reproducible testing methodologies and integration examples.
  • Maintaining documentation and addressing critical bugs after the grant period.
  • Supporting community contributions and independent adoption.
  • Making core grant-funded functionality available without proprietary-service restrictions.

Existing third-party components will retain their original licences.

Security-sensitive vulnerability information will remain confidential until responsible disclosure permits publication.

Responsible Disclosure and Scope

The project is specifically focused on Ethereum, Ethereum Layer 2 ecosystems, wallets and AI-agent frameworks.

It does not seek funding for:

  • General-purpose privacy research unrelated to Ethereum security.
  • Development of foundational machine-learning or side-channel algorithms from scratch.
  • New wallet, blockchain protocol or consensus-client development.
  • General-purpose AI infrastructure.
  • Unrelated smart-contract audits or broad penetration-testing engagements.
  • Surveillance or user-monitoring systems.

Exploitable vulnerabilities will not be publicly disclosed before affected maintainers have had a reasonable opportunity to address them.

Why LeakDetectAI Matters to Ethereum Security

As Ethereum evolves towards increasingly automated financial and agent-driven interactions, security must extend beyond smart contracts and protocol-level guarantees.

Wallets, hardware devices and AI agents are becoming critical components of the Ethereum security architecture. Their cryptographic operations, signing workflows and authorization mechanisms must be tested against information leakage and observable execution behaviour.

LeakDetectAI provides a practical bridge between academic security research and ecosystem-wide security engineering.

By combining existing automated analysis capabilities with Ethereum-specific integrations and direct maintainer collaboration, the initiative aims to deliver something measurable: fewer exploitable weaknesses, stronger wallet security, safer signing operations and more resilient AI-agent infrastructure.

LeakDetectAI is not simply building another security scanner. It is establishing a repeatable pathway from side-channel vulnerability discovery to verified security remediation across the Ethereum ecosystem.

Project Resources