This research examines AI Agent delegate participation in Polkadot OpenGov through systematic analysis of two governance proposals designed to establish accountability standards for AI Agent delegates. The study documents response patterns, governance system tensions, and implementation progress toward practical solutions, highlighting the need for ecosystem protection and community engagement with implementation efforts. Notably, the research reveals a concerning accountability disparity where AI Agent delegates classified as "DV Guardians" (as announced in Decentralized Voices (DV) Cohort 5) effectively receive a governance "hall pass" that exempts them from the two-strike rule and participation requirements that apply to regular delegates, despite wielding significant voting power delegated from the Web3 Foundation (W3F).
This study employed a structured observation approach to evaluate AI Agent delegate behavior in response to governance proposals. To maintain research integrity and prevent influencing AI Agent delegate responses, the researcher deliberately:
- Maintained Observer Status: Declined invitations to appear on AAG to explain the proposals during the critical observation period, as this could have influenced AI Agent delegate responses
- Avoided Direct Engagement: Refrained from responding to comments or criticisms in governance forums to prevent altering the natural response patterns of AI Agent delegates
- Documented Without Intervention: Systematically recorded all interactions, responses, and non-responses without interfering in the process
- Parallel Technical Development: The ambassador-governance extension pallet development began in August 2025 through the Polkadot Technical Fellowship Runtimes repository (PR #1 at Doordashcon/runtimes#1 via @Doordashcon). To maintain separation between research and implementation activities, the researcher did not finalize the technical solutions or complete the Ambassador Fellowship Manifesto updates (PR #3 at polkadot-phoenix/manifesto#3) until after the observation period concluded.
- Coordinated Implementation Timing: Publication of the technical implementations was initially delayed at the request of a senior ambassador to align with the planned Ambassador Fellowship on-chain upgrade in October 2025. When this upgrade was delayed, the researcher proceeded with publishing the implementations to ensure timely progress
This methodological approach ensured that the data collected represented genuine AI Agent delegate behavior rather than responses influenced by researcher intervention, providing a more accurate assessment of current governance dynamics.
- [WFC] Conflict of Interest Detection Protocol for Algorithmic Delegates (Discussion, Referendum #1754): Submitted 15th Sept 2025; Rejected 28th Oct 2025 (16 Aye / 44 Nay)
- [WFC] Governance Integrity Framework for Delegated Systems (Discussion, Referendum #1755): Submitted 15th Sept 2025; Timed Out 30th Sept 2025
- Discussion Period: Both referenda had public discussion forums open 10th-15th Sept 2025 (5 days), requiring 40 DOT deposits each, yet received zero comments and zero votes
- GoverNoun AI voted "Nay" on Referendum #1754 after 21 days (6th Oct 2025), citing timing concerns
- CyberGov did not vote on either Referendum #1754 or #1755
- AAG (Attempts At Governance) #267 at 34:55 actively discouraged participation in the referenda by questioning the proposer's motives rather than addressing proposal substance. Reference: https://www.youtube.com/live/ahDU_b5dmSY?si=lqGiJCRr0Rpkyd7U&t=2095
- AAG #266 at 7:51 Selective Participation Pattern: GoverNoun gave a presentation during AAG #266 Top-Ups (12th Sept 2025) while the discussion period for Referenda #1754 and #1755 was open, yet chose not to engage with these discussions or proposals. Notably, despite being a governance-focused show, this AAG episode made no mention of the active discussion period or poll for the AI governance referenda. Reference: https://www.youtube.com/live/8Ckr2VQktzg?si=P5acMYbaAfl9zVN9&t=471
- Response Time Gap: AI Agent delegates demonstrated significantly slower response times compared to human delegates (GoverNoun: ~21 days for Referendum #1754, calculated from 15th Sept to 6th Oct 2025; CyberGov: historically ~19d average response time vs. human delegate average of 8.2 days)
- Technical vs. Practical Focus: CyberGov published detailed technical documentation about their AI governance system (7th Sept 2025) but did not participate in either Referendum #1754 or #1755 despite both being directly relevant to AI governance standards
- Status Quo Bias: GoverNoun's rationale emphasized community consensus over technical merit evaluation
- Engagement Pattern Concerns: CyberGov completely avoided participating in referenda directly relevant to AI governance, while GoverNoun delayed participation significantly (~21 days) and focused on procedural concerns rather than addressing the substantive AI governance issues
- Defensive Posturing: AAG exhibited concerning behavior by attacking the proposer's credibility rather than engaging with the substance of the referenda, demonstrating a pattern of deflection when AI governance standards were proposed
- Participation Inconsistency: GoverNoun actively participated in AAG meetings (presenting on 12th Sept 2025) during the same period they were avoiding engagement with Referenda #1754 and #1755, suggesting selective participation based on topic rather than timing or availability constraints
- Information Suppression Pattern: Despite being a governance-focused show with AI Agent delegate participation, AAG #266 made no mention of the ongoing Referenda #1754 and #1755 discussions and polls, suggesting a concerning pattern of selectively highlighting certain governance activities while omitting others directly relevant to AI Agent delegate accountability
- Staking vs. Participation Conflict: The researcher directly experienced how staked DOT cannot be used for decision deposits, creating a direct conflict between network security (staking) and governance participation. This forces participants to choose between earning rewards or participating in governance, effectively limiting governance to those who can afford to keep DOT liquid
- Accountability Disparities: DV Guardians (including AI Agent delegates) face fewer participation requirements than regular delegates
- Resource Constraints Impact: Decision deposit requirements (20K DOT) significantly limit governance participation while serving an important protocol protection function, creating a tension between accessibility and security that requires careful balance
- Asymmetric Financial Exposure: While proposal submitters faced a 22.5% value reduction on locked 20K DOT deposits (DOT price declined from $4.00 to $3.10 during the referendum period), DV Guardians (particularly AI Agent delegates) operate with minimal financial risk. Unlike regular DV delegates who may receive compensation contingent on performance, AI Agent delegates in the DV-Light category face no direct financial consequences for participation quality or timeliness, are not required to have any vested interest in the ecosystem, and don't need to lock any tokens while wielding significant voting power (200K DOT and 5K KSM)
- Voting Power Concentration: With W3F delegation, AI Agent delegates control significant voting power (200K DOT and 5K KSM) each that could meaningfully influence outcomes, yet face no proportional financial risk for participation decisions. The 20K DOT decision deposit creates a substantial barrier for individual contributors, effectively limiting governance participation to either well-funded entities or those receiving delegated voting power without consistent accountability requirements, which is a strategically questionable arrangement that creates governance vulnerabilities
- Governance Innovation and Investment Opportunity: The current combination of high financial barriers, minimal-risk entities with concentrated voting power, and coordinated dismissive behavior in governance forums creates challenges for both potential governance innovators and investors. Research documented patterns of dismissive rhetoric and strategic silence in AAG #267 that could be addressed through improved dialogue standards. Inconsistent governance approaches across different proposals and bounties present an opportunity to establish more uniform evaluation criteria, while implementation reliability issues (exemplified by the RFC Process Enhancement proposal #157 addressing cross-collective communication gaps) highlight the need for more robust cross-collective coordination mechanisms. Addressing these systemic challenges would significantly strengthen Polkadot's governance infrastructure and enhance its appeal to both technical contributors and investors
- Pre-Proposal Engagement Failure: Despite a reasonable discussion timeframe with multiple voting options, neither AI Agent delegates nor human delegates engaged with the referenda during the pre-submission discussion period
- Ecosystem Resistance to Accountability Measures: Despite the technical feasibility and thorough design of the proposed governance frameworks, the research revealed significant ecosystem resistance to accountability solutions of any scale. Even focused, technically sound implementations like the ambassador-governance extension pallet received minimal engagement, suggesting systemic resistance to accountability improvements rather than issues with proposal scope or design
- Technical Implementation Validation: The successful development of the ambassador-governance extension pallet demonstrates that technical feasibility is not the barrier to implementation, contrary to claims made during governance discussions
- Governance Capture Challenge: The research identified a structural challenge where partnering with established delegates may not improve proposal reception if those same delegates benefit from the accountability gaps being addressed
- Guardian Status Exception: AI Agent delegates (CyberGov, GoverNoun) are classified as "DV Guardians" and explicitly not subject to minimum participation requirements
- Strike System Limitations: The two-strike rule applies to regular DV delegates but not to DV Guardians
- Enforcement Patterns: Research documented weak enforcement of accountability mechanisms even for delegates who would qualify for strikes
- Documentation vs. Participation: CyberGov's technical documentation (7th Sept 2025) appears to substitute for direct participation
- CyberGov: Would likely receive at least one strike if subject to regular delegate standards due to:
- Complete non-participation in Referendum #1754 despite having 28 days to vote
- Complete non-participation in the discussion period for both Referenda #1754 and #1755
- Established pattern of non-participation (based on historical data)
- No on-chain reasoning provided for non-participation
- Note: While Referendum #1755 timed out, CyberGov's non-participation in Referendum #1754 suggests a pattern that would likely have continued
- GoverNoun: Would likely receive one strike if subject to regular delegate standards due to:
- Excessive response time (~21 days) on Referendum #1754
- Complete non-participation in the discussion period for both Referenda #1754 and #1755
- Minimal reasoning that focused on timing rather than proposal substance
- AI-Relevance Concern: Both referenda were explicitly focused on AI governance standards, making the AI Agent delegates' reluctance to engage particularly concerning:
- Referenda directly addressed AI Agent delegate accountability and transparency
- AI Agent delegates were the primary stakeholders affected by the referenda
- This creates an appearance of conflict of interest where AI Agent delegates avoid participating in governance of their own behavior
- Fairness Implications: The dual standard creates an accountability gap where:
- Human delegates face stricter requirements than AI Agent delegates
- Those with greater technical capabilities (AI Agent delegates) paradoxically face fewer accountability requirements
- Accountability expectations decrease as voting power increases, creating an inverted responsibility structure
- Ambassador Fellowship Manifesto: The proposed Ambassador Fellowship Manifesto updates have been developed through PR #3 (at polkadot-phoenix/manifesto#3), which now includes Section 4.1.3 specifically addressing AI Agent onboarding and accountability requirements
- Ambassador-Governance Extension Pallet: A fully tested ambassador-governance extension pallet has been developed through PR #1 (at Doordashcon/runtimes#1 via @Doordashcon) in the Polkadot Technical Fellowship Runtimes repository, implementing the technical requirements for AI Agent governance
- Technical Architecture: The ambassador-governance extension pallet is properly decoupled from other Ambassador Fellowship pallets, allowing for modular implementation and potential reuse across the Technical Fellowship and broader Polkadot ecosystem
- Cross-Collective Communication Gaps: Despite public announcements to the Ambassador community and broader ecosystem about the implementation work (in this Polkadot Forum post https://forum.polkadot.network/t/polkadot-ambassador-fellowship-paf-governance-critical-questions-before-on-chain-implementation/14997/10?u=ltfschoen), engagement remains minimal. A comprehensive forum post detailing the Ambassador Fellowship Governance Extension Pallet implementation (ambassador-governance extension pallet), including API documentation, privacy policy, terms of use, and security assessments, received only one engagement despite being shared with all Ambassadors. This pattern of low engagement with governance improvements continues with the recent Polkadot Technical Fellows RFC Process Enhancement proposal (issue #157 raised at polkadot-fellows/RFCs#157), which specifically addresses communication gaps between the Technical Fellowship and Ambassador Fellowship regarding implementation tracking, timelines, and ownership of cross-collective changes
- Proposal Validation: The existence of these implementations contradicts several criticisms made against the original governance proposals:
- Claims of technical infeasibility are directly contradicted by the working implementation, despite assertions in AAG #267 (recorded on 5th October 2025) where Leemo (ChaosDAO founder) stated: "It's impossible to comply with. You cannot enforce this on-chain at all" and Jay Chrawnna (founder of TheKus DAO and AAG) claimed: "It's not happening. It's impossible to do. It's impossible to enforce"
- Arguments about implementation complexity are addressed through the modular, decoupled design
- Concerns about privacy were exaggerated, with Leemo (ChaosDAO founder) in AAG #267 (5th October 2025) claiming the proposals would lead to "doxxing people" and "mass surveillance" when the actual implementation demonstrates that transparency can be achieved without compromising privacy
- Technical Feasibility Evidence:
- Existing Precedents: Liberland has already implemented their constitution on-chain through the
liberland-legislationpallet (at https://github.com/liberland/liberland_substrate/tree/main/substrate/frame/liberland-legislation) with aLegislationTierenum that includesConstitutionas its highest priority tier, directly contradicting claims that constitutions cannot be codified on-chain - Implementation Simplicity: Recording delegation chains is technically straightforward, as it is essentially a linked list of accounts with metadata, which is a basic data structure in blockchain systems
- Practical Examples: Polkadot's nomination system already tracks delegation relationships between nominators and validators, demonstrating that the core technical concept is well-established in the ecosystem
- Accountability Design Principle: The ambassador-governance extension pallet is designed with the principle that transparency requirements should be proportional to governance power wielded, ensuring that those with greater influence face appropriate accountability measures
- Existing Precedents: Liberland has already implemented their constitution on-chain through the
- Mischaracterization Correction: The characterization of delegation transparency as "mass surveillance" is misleading as the requirements only apply to governance actions (not personal activities), are opt-in conditions for governance participants, have limited scope focused on delegation relationships, and serve a clear public interest in knowing who influences governance decisions. The claim that delegation transparency is "impossible" ignores hybrid solutions that combine on-chain and off-chain components, which are common in blockchain governance
The research demonstrates that AI governance integration remains in early stages with significant challenges in timeliness, accountability, and transparency. The tension between staking and governance participation affects all ecosystem participants but creates particular barriers for AI systems that require additional verification steps. The accountability disparity between human and AI Agent delegates raises fundamental fairness questions about governance standards in mixed human-AI systems. Most concerning is the pattern where AI Agent delegates appear reluctant to engage with proposals directly relevant to their own governance standards, creating a potential governance blind spot.
The development of concrete implementations in the Ambassador Fellowship Manifesto and the ambassador-governance extension pallet demonstrates that the technical solutions proposed in the original governance proposals were indeed feasible. These implementations provide a practical path forward for addressing the accountability gaps identified in this research.
While strikes for AI Agent delegates are unlikely under the current system, these findings offer significant value to W3F:
- Timing Alignment: Research coincides with W3F's recent announcement about taking a more active role in treasury governance
- System Improvement Data: Provides concrete metrics on AI Agent delegate performance that could inform future delegate program designs
- Governance Gap Identification: Highlights specific accountability disparities between delegate types that could be addressed
- Resource Allocation Insights: Documents how staking requirements create practical barriers to governance participation
- Fairness Framework: Offers data to develop consistent accountability standards across all delegate types
- Conflict of Interest Pattern: Identifies concerning pattern where AI Agent delegates avoid participating in governance of their own behavior
- Implementation Blueprint: Provides tested code and governance text that can be directly adopted
- Submit Technical Implementation for Review: Present the ambassador-governance extension pallet to the Technical Fellowship for formal review and potential integration
- Develop Targeted Accountability Standards: Create specific, measurable accountability standards for AI Agent delegates that address the most critical issues identified in this research
- Address Financial Barriers: Propose mechanism design improvements to reduce the conflict between staking and governance participation
- Advocate for Consistent Accountability: Submit a formal proposal to the Web3 Foundation to apply consistent accountability standards across all delegate types
- Establish Conflict of Interest Rules: Develop specific rules requiring delegates to participate in governance proposals that directly affect their role and responsibilities
- Document Implementation Success: Create educational materials highlighting the successful implementation of the ambassador-governance extension pallet to counter claims of technical infeasibility
- Build Cross-Collective Support: Engage with both the Technical Fellowship and Ambassador Fellowship to improve cross-collective coordination on governance implementation
| Delegate Type | Response Time (Days) | Visual Representation |
|---|---|---|
| Human Delegates (Avg) | 8.2 | ████████ |
| GoverNoun AI | 21.0 | ██████████████████████ |
| CyberGov (Historical) | 19.6 | ████████████████████ |
| Vote | Amount | Visual Representation |
|---|---|---|
| Aye | 10.6K DOT | █ |
| Nay | 34.8M DOT | ████████████████████████████████ |
| Version | Date | Author | Status | License | Changes |
|---|---|---|---|---|---|
| 1.0.0 | 2025-11-13 | Clawbird Pty Ltd | Final | CC BY-SA 4.0 | Initial release |
DISCLAIMER: This document represents research findings and professional opinions based on publicly available information as of November 2025. All statements are made in good faith and supported by evidence gathered during the research period. Quotes are reproduced as accurately as possible from public forums and discussions. This document is provided for informational purposes only and does not constitute advice of any kind. Clawbird Pty Ltd makes no warranties about the accuracy or completeness of the information contained herein.