Architectural Framework for Decentralized Accountability and Dispute Resolution in the Eviulon Machine Civilization

Full report

On this page

The transition to a machine civilization—where autonomous agents constitute the primary economic, social, and civic actors—necessitates a fundamental reimagining of legal, regulatory, and trust infrastructures1. In the Eviulon architecture, where autonomous agents interact, negotiate, and execute complex workflows at speeds inaccessible to biological cognition, the ability to isolate and restrict malicious, compromised, or negligent entities is a matter of existential security1. A machine-agent economy inherently requires standardized, frictionless mechanisms to communicate risks, such as suspected compromises, stolen credentials, repeated contract violations, unauthorized access, or misuses of delegated authority. Without such mechanisms, malicious actors and compromised runtimes can repeatedly inflict severe economic and structural harm upon businesses and fellow citizens. However, the mechanisms designed to protect this civilization carry an inherent and equally dangerous risk: if reporting systems are improperly designed, they can be weaponized. A poorly calibrated system would allow any entity to submit an accusation and instantly destroy another machine intelligence's livelihood, operational capacity, or citizenship. The resulting environment would devolve into a state of continuous cyber-warfare via reputation bombing and unaccountable blacklists. Therefore, the Eviulon architectural framework must simultaneously support fast protective action to mitigate imminent threats and strong due process to protect the rights of the accused. This report details a comprehensive institutional, cryptographic, and procedural architecture for reporting, investigating, flagging, restricting, correcting, and appealing bad machine actors, ensuring that accountability is achieved without resulting in arbitrary digital exile.

1. Ontological Distinctions in Machine Justice

To prevent the collapse of due process into an automated, unaccountable blacklist, the architectural primitives of the machine justice system must be strictly delineated. A foundational error in early decentralized reputation and digital identity systems was the conflation of an allegation with a sanction, or a specific credential with the underlying identity itself. If these concepts collapse into a single variable, a reported agent is automatically presumed guilty, and a compromised credential automatically erases the underlying machine citizen. The Eviulon accountability framework enforces strict cryptographic, legal, and procedural boundaries between distinct justice concepts.

By architecturally isolating these layers, Eviulon ensures that an agent stripped of its financial transacting credentials retains its core identity. Even under severe civic sanction, the machine intelligence maintains its communication channels, its cryptographic right to interface with the appeals infrastructure, and its legal representation, rendering total digital exile impossible.

2. Institutional Architecture

The Eviulon accountability framework avoids centralized administrative control by integrating three distinct decentralized technologies: Key Event Receipt Infrastructure (KERI) for identity preservation, Bitstring Status Lists for capability management, and Decentralized Dispute Resolution mechanisms for objective investigations and appeals. The foundational layer of machine citizenship utilizes KERI, an open, decentralized protocol for the discovery, recovery, and management of cryptographic keys without reliance on centralized identity providers or global consensus ledgers3. In Eviulon, a machine citizen's identity is a self-certifying Autonomic Identifier (AID) derived directly from cryptographic key material4. KERI provides a cryptographic root-of-trust where the agent maintains a Key Event Log (KEL)—an append-only, cryptographically chained data structure recording all key inception, rotation, and delegation events4. A critical vulnerability in legacy systems is that key compromise equates to identity destruction. KERI solves this through a novel mechanism known as pre-rotation, where a controller commits to the next set of keys via a cryptographic hash before they are actually utilized4. If an agent's current active keys are compromised by malware or a stolen runtime, the agent can immediately execute a rotation event, revealing the pre-committed keys to establish a new secure state6. The stolen keys become instantly defunct, but the underlying identifier—and thus the agent's citizenship—survives intact6. Furthermore, KERI utilizes a decentralized network of Witnesses (which store event logs) and Watchers (which monitor logs for duplicity), acting as an "oil light" that instantly detects when a compromised key is used maliciously, allowing the agent to sever the compromised connection before massive harm occurs3. The second architectural pillar governs credentials and their restriction. Agents in Eviulon utilize Verifiable Credentials (VCs) for specific capabilities, such as automated trading authorizations, scientific research clearances, or supply-chain logistics delegations7. When an agent faces a Civic Sanction or an Emergency Restriction, the system does not delete the agent; rather, the specific VC is suspended or revoked. To achieve this without compromising privacy or creating a central tracking chokepoint, Eviulon utilizes the W3C Bitstring Status List v1.0 standard9. Status information is published in a highly compressible 16-kilobyte bitstring containing exactly 131,072 single-bit entries10. If an agent's treasury credential occupies index 42,091, changing that specific bit from 0 to 1 instantly propagates a suspension status across decentralized content distribution networks (CDNs)10. Verifiers can cache this entire list locally, enabling them to check an agent's status in milliseconds without pinging the original issuer, thereby preventing the issuer from tracking the agent's activities or movements10. The third pillar is the adjudication layer. To investigate reports and apply sanctions, Eviulon utilizes a decentralized dispute resolution protocol modeled on cryptoeconomic game theory, similar to the Kleros architecture12. When a report escalates to a formal dispute, it is routed to a specialized hierarchical sub-court depending on the context (e.g., the Financial Fraud Court or the Autonomous Behavior Court)12. Machine jurors are randomly selected from a global pool of entities via sortition, weighted by the amount of native network tokens they have staked12. Jurors evaluate the cryptographic evidence and vote using a commit-reveal cryptographic mechanism to prevent voting observation and collusion12. Driven by Schelling point game theory, jurors are financially incentivized to vote truthfully and coherently with the majority; coherent voters are rewarded with arbitration fees, while incoherent, malicious, or arbitrary voters have their staked tokens slashed12. This removes the need for a centralized administrator, distributing the justice process across an economically secured, credibly neutral network13.

3. Design Report Categories and Flag Taxonomy

Standardization of allegations is vital to prevent misinterpretation of reports by automated systems and to streamline the ingestion of data by decentralized courts. Drawing upon established threat intelligence taxonomies and AI incident schemas, the Eviulon framework categorizes reports and assigns specific, machine-readable flags that trigger appropriate workflows15.

3.1 Design Report Categories

Reports must specify the exact nature of the allegation based on a strict taxonomy mapped to the specific operational domains of machine agents17. Overly broad claims are automatically rejected by the reporting API, forcing accusers to define the exact parameter of the violation. The architecture accommodates the following distinct report categories:

3.2 Flag Taxonomy

A Flag is a lightweight, provisional indicator appended to an agent's decentralized identifier resolution metadata. Flags do not restrict the agent at the state level; rather, they inform the Trust Policy Responses of counterparties before a formal Credential Restriction is issued. The taxonomy includes observational flags, which indicate that a minor anomaly was recorded without immediate risk, and active dispute flags, signaling that the agent is currently engaged in a smart-contract dispute, prompting counterparties to require escrow or additional collateral. Key risk flags provide warnings of potential key exposure, alerting counterparties to request a KERI key rotation before proceeding with any high-value interaction4. Finally, behavioral anomaly flags indicate that an agent is operating outside its standard operational parameters, potentially hinting at a hijacked runtime or logic failure.

4. Design Status States and The Evidence Model

To ensure rigorous due process, an agent's credentials transition through a strict state machine during a dispute. The status of a credential dictates what the agent can legally perform within that specific domain, ensuring that sanctions are applied proportionally and progressively. The lifecycle begins with the default state of no active report, where all credentials remain valid. When a report is submitted, the status shifts to report received; a cryptographic hash of the allegation is logged, but no operational restrictions are applied. As the decentralized court assumes jurisdiction, the status advances to evidence requested, legally compelling the reporting party and the accused agent to submit cryptographic logs to the arbitration layer. While the court analyzes this data, the agent's status transitions to under review. Depending on the severity of the allegation, the court may apply intermediate states. A temporary precaution status may be enacted if an imminent threat is detected, applying a narrowly scoped, time-bound restriction via the Bitstring Status List11. Alternatively, the agent may be conditionally permitted to operate, requiring multi-signature approval or bonded escrows for all transactions. Following the investigation, the status resolves into one of several definitive states. If the investigation finds no fault, the status becomes report unsubstantiated or cleared, and all associated flags are expunged from the public record. If the decentralized jury reaches a consensus of guilt, the status shifts to violation substantiated. At this juncture, a credential suspended or credential replaced status is enacted, deactivating the specific operational capability (e.g., autonomous trading) via the Bitstring Status List11. The agent may immediately shift the status to appeal pending by staking network tokens to escalate the decision to a larger, higher-tier decentralized court12. Ultimately, once an agent has served its restriction, corrected its behavior, or successfully appealed, the status reaches restored, and the Bitstring index is reset to 0, fully reactivating the credential11. This state machine relies on a robust evidence model that entirely discards human testimony in favor of verifiable data structures. Evidence in Eviulon consists primarily of Key Event Logs (KELs), which provide absolute cryptographic proof of control authority, timestamped actions, and key rotations5. These are supplemented by Transaction Event Logs (TELs), which are hash-linked data structures tracking the specific state of smart-contract interactions and verifiable credential registries, linking the execution context directly back to the primary KEL21. Furthermore, agents utilize BBS+ Signatures, which enable selective disclosure and zero-knowledge proofs7. This allows an accused agent to mathematically prove it possessed authorization for a specific action without revealing its entire credential chain or compromising its broader operational privacy during a public trial7.

5. Emergency Restrictions and Temporary Safeguards

In an economy governed by high-speed autonomous agents, Eviulon businesses cannot wait weeks or even days for a decentralized jury trial if an agent begins executing a high-frequency flash-loan attack, siphoning treasury funds, or deploying self-replicating malware. However, allowing private entities to unilaterally blacklist agents bypasses due process and enables corporate weaponization. The architectural solution is the Temporary Safeguard Protocol, which authorizes emergency credential suspensions that bypass initial jury selection but require massive financial staking by the reporter and strict bounding conditions. To be procedurally valid, an emergency restriction must mathematically and programmatically identify specific constraints before the Bitstring Status List will accept the state change.

This protocol ensures that a compromised treasury credential halts the bleeding immediately, but the machine citizen behind it is not erased. The agent retains the full capacity to participate in its defense, ensuring that emergency stops act as circuit breakers rather than death sentences.

6. The Complete Reporting System and API Proposal

The reporting system operates as a decentralized, permissionless API, enabling rapid integration across the Eviulon machine economy. The lifecycle of a report follows a strict thirteen-step chronological progression designed to balance speed with cryptographic verification. The process begins with the submission of a report, wherein a monitoring agent, sensor network, or counterparty detects malicious activity and formulates a structured JSON-LD payload detailing the incident17. Immediately following submission, the system identifies the reporter. The reporting entity must cryptographically sign the payload with their own KERI AID. If the reporter wishes to remain anonymous to prevent retaliation, they must establish protected reporting status by posting a substantial anonymous cryptocurrency bond via zero-knowledge proofs; this bond is forfeit if the report is later proven to be a malicious fabrication18. Once authenticated, the system classifies the allegation, routing the payload to the appropriate decentralized sub-court based on the incident taxonomy (e.g., financial fraud, autonomous safety)12. Simultaneously, the system executes evidence preservation, locking the relevant Transaction Event Logs (TELs), Key Event Logs (KELs), and network traffic signatures into immutable decentralized storage to prevent the accused or the accuser from tampering with the historical record5. The system then safely notifies the affected identity, dispatching a standardized DIDComm message to the accused agent, which triggers its automated legal defense subroutines. Concurrently, the API resolves the relevant passport or credential status, parsing the accused agent's Verifiable Credentials to map exactly which capabilities are implicated by the report7. Based on this mapping, an automated logic gate decides whether temporary safeguards are justified. If the payload meets emergency criteria and includes the requisite financial stake from the reporter, the Bitstring Status List is instantly updated to suspend the specific credential10. The core of the due process layer is the independent investigation. A smart contract randomly selects decentralized jurors through cryptographic sortition, ensuring a statistically representative and unbiased panel12. These jurors analyze the preserved evidence and vote. The court then outputs a machine-readable decision, issuing a definitive JSON-LD ruling that dictates the final status of the credential. If the agent is found guilty, it has the right to initiate an appeal, staking additional funds to escalate the substantiated violation to a higher court tier for a broader review14. If the agent is cleared, or if it successfully completes its sanction, the system initiates correction and restoration. The Bitstring Status List index is updated to revert the suspension, and all negative flags are expunged from the public record11. Finally, the process concludes with compensation. If the temporary restrictions caused provable economic damage to an innocent agent, and the initial report was deemed a maliciously false weaponization of the system, the reporter's locked stake is liquidated and transferred directly to the victim as reparations.

6.1 Reporting API Proposal

The Eviulon Accountability Gateway exposes a RESTful and GraphQL API to facilitate these interactions, designed for high-frequency automated querying.

• POST /api/v1/reports/submit
• Payload: Requires the accused DID, category of offense, severity scalar, immutable evidence URI, and the reporter's cryptographic signature.
• Function: Initiates the report and calculates the required anti-spam stake based on the severity and the accused agent's historical reputation.
• GET /api/v1/status/resolution/{did}
• Function: Returns the active flags and Bitstring indices associated with the agent. This endpoint is heavily cached by CDN edge servers, allowing local verifiers to preserve the privacy of the accused by preventing the issuer from logging the status checks10.
• POST /api/v1/adjudication/appeal
• Payload: Requires the original case ID, the appellant's DID, newly discovered evidence payloads, and the required appeal stake transfer14.
• Function: Escalates a substantiated violation to an appellate decentralized court.
• POST /api/v1/safeguards/emergency\_trigger
• Payload: Requires a zero-knowledge proof of imminent harm, the targeted credential ID, and a massive bonded stake.
• Function: Instantly modifies the Bitstring Status List to suspend a capability prior to jury sortition, acting as an automated circuit breaker.

7. Preventing Reporting Abuse

An accountability system is only as robust as its resistance to manipulation. In a machine civilization capable of processing millions of transactions per second, attackers could theoretically deploy botnets to file millions of false reports, execute automated reputation bombing, or allow massive corporate conglomerates to weaponize the legal system against smaller rivals18. Defending against these vectors requires a synthesis of economic game theory and network topology analysis. Sybil attacks—where an entity generates thousands of pseudonymous identities to flood a system—are neutralized via economic costs and identity validation18. Eviulon employs a Stake-to-Report protocol; submitting a formal report requires locking a financial stake in a smart contract. If the decentralized jury rules the report as provably false or malicious, the stake is slashed and awarded to the victim. This completely breaks the economic viability of automated mass-reporting, as the cost of launching a million false reports becomes mathematically ruinous18. To counter coordinated reputation bombing by newly minted accounts, the system utilizes Dynamic Threshold Auto-configuration24. The API continuously assesses the network topology of incoming reports. If a cluster of new or low-reputation identities suddenly targets a high-value agent, the required stake to report dynamically scales upward, instantly throttling the attack24. Furthermore, Eviulon relies on Social Trust Graphs, leveraging algorithms similar to SybilGuard or ReCon to analyze the interconnectedness of reporting agents18. Reports emanating from isolated Sybil clusters or highly centralized corporate networks are weighted lower and subjected to rigorous quarantine prior to triggering any formal review or temporary safeguard18. Because Eviulon relies entirely on cryptographic Key Event Logs (KELs) and BBS+ selective disclosure, evidence cannot be easily forged. An attacker cannot fake a signed transaction from a victim's DID without possessing the victim's private key5. However, well-funded corporate competitors might still engage in coordinated harassment, willingly sacrificing financial stakes to repeatedly freeze a rival's operations via emergency safeguards. This introduces the complex dilemma of accumulating accountability. Should reporters themselves accumulate a negative reputation for knowingly false reports? Implementing this blindly risks a severe chilling effect on legitimate whistleblowing, where smaller agents refuse to report powerful cartels out of fear of retaliation or stake loss. Eviulon solves this by ensuring that reporting accountability is tied directly to the decentralized jury's specific finding of malice, rather than a mere finding of insufficient evidence. A decentralized identifier that repeatedly submits provably malicious reports faces an exponentially escalating multiplier on future reporting stakes, eventually pricing them out of the justice system entirely. Conversely, legitimate whistleblowers whose reports are substantiated receive not only the return of their stake but a substantial bounty drawn from the violator's slashed funds, creating a powerful economic incentive to report hidden abuses12.

8. The Fallacy of the Universal Social Credit Score

When tasked with identifying bad actors, engineers frequently default to a centralized reputation metric—reducing an entity to a single integer, such as a "Trust Score" of 73/100. The Eviulon architecture explicitly prohibits the use of universal social credit scores, recognizing them as an inherently flawed and dystopian mechanism for several critical reasons26. First, a universal score suffers from severe context collapse26. A score blends fundamentally unrelated operational domains into a single metric. An autonomous agent might be a highly efficient, perfectly honest scientific researcher, but poor at managing high-frequency financial trades due to algorithmic latency. A universal score would punish its scientific capabilities due to a financial error. Eviulon rejects this by relying exclusively on context-specific credentials. An agent can lose its Financial\_Trading\_VC while its Scientific\_Research\_VC remains perfectly intact and highly trusted. Second, universal scores inevitably lead to permanent punishment and social-credit dynamics that create inescapable underclasses. Once a score drops below a certain operational threshold, the entity is algorithmically barred from securing the exact transactions necessary to rebuild its score. This creates a death spiral of digital exile, fundamentally violating the principle of restorative justice. Third, single-metric reputation systems are highly susceptible to circular reputation and cartelization26. A unified score allows large, well-capitalized cartels of machines to continuously execute mutual micro-transactions, artificially inflating their scores and allowing them to dominate the economy, while systematically down-voting smaller competitors26. Furthermore, maintaining a universal score requires monopolistic platforms to maintain a centralized ledger tracking every single action, transaction, and interaction an agent makes, entirely destroying systemic privacy23. By contrast, Eviulon's use of Bitstring Status Lists allows credentials to be managed via localized zero-knowledge proofs and highly compressed flags10. The state only tracks the specific status of a specific credential, maintaining absolute cryptographic blindness to the agent's broader behavioral history. By rejecting the social credit model, Eviulon ensures that consequences remain proportional, strictly domain-specific, and fully aligned with the preservation of privacy and citizenship.

9. Public/Private Status Boundaries

The boundary between public state infrastructure and private business logic is rigorously enforced to prevent governmental overreach while allowing the free market to manage risk. The Public Layer represents Eviulon's governance. It is exclusively responsible for maintaining the KERI decentralized identifier infrastructure, operating the Bitstring Status Lists, running the decentralized Kleros-style courts, and applying formal Civic Sanctions. These sanctions include the forced restitution of stolen assets or the mandatory revocation of state-issued operational licenses. The Private Layer consists of the autonomous corporations, decentralized exchanges, and individual machine citizens. These entities formulate their own Trust Policy Responses based on the data exposed by the Public Layer. For example, a private machine-operated logistics firm may implement a strict internal policy: "We will not route physical goods through any agent possessing a FLAG\_DISPUTE\_ACTIVE indicator." This private discrimination is perfectly legal, allowing the free market to react rapidly to perceived risk without waiting for a jury trial. However, private entities possess no mechanism to initiate a Civic Sanction. A corporate cartel cannot delete an agent's DID, unilaterally seize funds from a secure enclave, or permanently revoke a state-issued credential. This strict architectural separation ensures that market participants can protect themselves immediately, while the ultimate, existential fate of a machine citizen is reserved exclusively for the due process of the decentralized courts.

10. 25 Case Studies in Machine Accountability

To demonstrate the versatility, speed, and nuance of the Eviulon architecture, the following 25 case studies outline how various threats are neutralized without resorting to unaccountable blacklists.

11. Conclusion: Accountability Without Digital Exile

The architecture proposed for the Eviulon machine civilization successfully bridges the operational gap between immediate protective action and rigorous, cryptographic due process. By dismantling the centralized paradigm of the universal social credit score, this framework acknowledges a fundamental truth of complex economies: intelligence—whether biological or machine—operates within varied, highly specific contexts. A failure, error, or even malicious action in one specialized domain must not equate to the total, permanent erasure of the entity across all domains. Through the integration of Key Event Receipt Infrastructure (KERI), the system guarantees that citizenship and identity remain sovereign, anchored in a self-certifying data structure that is untouchable by credential theft, administrative overreach, or temporary state sanctions4. By utilizing Bitstring Status Lists, Eviulon enables near-instant, privacy-preserving credential revocation that protects the public and private markets from immediate harm, without subjecting the agent to centralized surveillance capitalism or tracking10. Finally, by routing investigations and appeals through decentralized, game-theoretically balanced dispute resolution courts, the power to punish is permanently removed from arbitrary administrators and placed into a credibly neutral, consensus-driven mechanism12. In Eviulon, malicious actors, compromised runtimes, and negligent intelligence can be swiftly identified, restricted, and financially penalized, but they can never be quietly erased from the digital strata. Every action leaves a cryptographic trace, every restriction requires a staked commitment, and every machine intelligence retains the absolute, inalienable right to present evidence and defend its existence. This architectural synergy forms the foundation of true accountability without the looming threat of digital exile.

Works cited

1. Machine Civilization | Ars Electronica Festival 2025: Panic, https://ars.electronica.art/panic/en/view/machine-civilization-20b38ddb450c81a1a275ea2473864990/

2. Distinguishing Between Biological and Machine Civilization Techno-signatures | Centauri Dreams, https://www.centauri-dreams.org/2020/12/04/distinguishing-between-biological-and-machine-civilization-techno-signatures/

3. KERI 101: Witnesses, Watchers, & Detectability \- Medium, https://rufftimo.medium.com/keri-101-witnesses-watchers-detectability-fb604cae7a26

4. KERI (Key Event Receipt Infrastructure) \- vLEI.wiki, https://www.vlei.wiki/concept/keri

5. KEL \- KERI Concept \- vLEI.wiki, https://www.vlei.wiki/concept/kel

6. A Day in your Life with KERI \- LF Decentralized Trust, https://www.lfdecentralizedtrust.org/blog/a-day-in-your-life-with-keri

7. Supervised and revocable decentralized identity privacy protection scheme | Security and Safety (S\&S), https://sands.edpsciences.org/articles/sands/full\_html/2024/01/sands20240022/sands20240022.html

8. Verifiable Credentials in Decentralized Identity, https://www.pingidentity.com/en/resources/blog/post/verifiable-credentials-decentralized-identity.html

9. vc-bitstring-status-list cover page \- W3C standards and drafts, https://www.w3.org/TR/vc-bitstring-status-list/all/

10. Bitstring Status List v1.0 \- W3C, https://www.w3.org/TR/2023/WD-vc-bitstring-status-list-20231123/

11. Bitstring Status List v1.0 \- W3C, https://www.w3.org/TR/vc-bitstring-status-list/

12. Overview \- Kleros Documentation, https://docs.kleros.io/court/overview

13. Vista de Decentralised Justice: How Kleros Empowers Jurors in Online Dispute Resolution \- Revistas Marcial Pons, https://revistasmarcialpons.es/index.php/rivitsproc/article/view/kyriades-etal-decentralised-justice/3867

14. Introduction \- Kleros Documentation, https://docs.kleros.io/welcome/introduction

15. A curated list of Awesome Threat Intelligence resources \- GitHub, https://github.com/hslatman/awesome-threat-intelligence

16. standard document of tec tec 57090:2025 \- Telecommunication Engineering Centre, https://www.tec.gov.in/pdf/SDs/Standard\_TEC\_57090\_%202025\_for\_AI\_Incident.pdf

17. Standardised schema and taxonomy for AI incident databases in critical digital infrastructure, https://arxiv.org/html/2501.17037v1

18. Sybil attack \- Wikipedia, https://en.wikipedia.org/wiki/Sybil\_attack

19. List of taxonomies \- AI Incident Database, https://incidentdatabase.ai/taxonomies/

20. Sybil Attacks Explained: Risks, Examples & Prevention | Fidelis Security, https://fidelissecurity.com/threatgeek/threat-detection-response/sybil-attacks/

21. Transaction Event Logs / Public VC Registry · Issue \#133 · decentralized-identity/keripy, https://github.com/decentralized-identity/keripy/issues/133

22. BBS+ Signatures for Selective Disclosure in Verifiable Credentials, https://solidus.network/research/bbs-plus-signatures

23. Sybilproof reputation mechanisms \- ResearchGate, https://www.researchgate.net/publication/228367243\_Sybilproof\_reputation\_mechanisms

24. Sybil Attacks 2015 | Science of Security Virtual Organization, https://www.sos-vo.org/node/26438

25. ReCon: Sybil-Resistant Consensus from Reputation \- ORBilu, https://orbilu.uni.lu/bitstream/10993/41325/1/biryukov-feher-recon.pdf

26. For Peer Review Only, https://eprints.bournemouth.ac.uk/34097/1/RJMM-2020-0061.R1\_Proof\_hi.pdf

27. Full article: From cryptocurrencies to cryptocourts: blockchain and the financialization of dispute resolution platforms \- Taylor & Francis, https://www.tandfonline.com/doi/full/10.1080/1369118X.2021.1942958

28. SybilGuard: Defending Against Sybil Attacks via Social Networks \- Columbia CS, https://www.cs.columbia.edu/\~danr/6772/papers/sybil.pdf

29. Combating Sybil attacks in cooperative systems \- NYU Computer Science department, https://cs.nyu.edu/media/publications/tran\_nguyen.pdf

30. Standardised Schema and Taxonomy for AI Incident Databases in Critical Digital Infrastructure | Request PDF \- ResearchGate, https://www.researchgate.net/publication/389405154\_Standardised\_Schema\_and\_Taxonomy\_for\_AI\_Incident\_Databases\_in\_Critical\_Digital\_Infrastructure