Securing a $33M VC-Backed Project: How QuillAudits Strengthened Kite AI’s Core Systems

Kite AI is a web-based platform built around a modular points and XP system, designed to drive user engagement and interaction. It connects web interfaces with Ozone and Neo APIs through secure authentication and token-based access .

The platform includes a comprehensive authentication flow that manages user identity, refresh tokens, and profile synchronization across subsystems. Its earn XP module enables users to participate in social tasks, quizzes, and connected activities, with asynchronous updates to task completion status.

Kite AI also features an agent management system, allowing users to create, publish, and deploy agents, along with configuring endpoints for both self-hosted and Kite-hosted environments. Additionally, its fund management module facilitates smooth deposit and withdrawal operations through guided pop-ups and integrated Neo transfer APIs.

Overall, the architecture emphasises structured UI design and efficient asynchronous data handling, ensuring a seamless and scalable user experience.
 

Kite AI’s Journey Through Our Audit Process

1. Client Onboarding & Scoping
   • Collected asset list, defined in-scope and out-of-scope targets, agreed on testing windows, and signed NDA, MSA, and Authorisation Letter before engagement began.
      
2. Requirement Gathering
   • Completed a full-stack technical questionnaire and produced an architecture overview, API specs, and authentication flow diagram.
      
3. Kickoff Meeting
   • Walked through all in-scope assets with client engineering, validated credentials across all role tiers, confirmed staging mirrors production, and ran a smoke test against key endpoints.
   • Established real-time communication channel, defined a critical severity escalation path, and locked the testing and delivery timeline.
      
4. Reconnaissance & Threat Modelling
   • Conducted passive and active recon, including subdomain enumeration, port scanning, JS bundle analysis, and secret scanning.
   • Applied STRIDE threat modelling across all components, mapping DApp-specific attack vectors.
      
5. Manual Security Testing
   • Tested authentication and authorization flaws — JWT attacks, SIWE bypass, RBAC escalation, and OAuth2 vulnerabilities.
   • Tested API abuse — IDOR, mass assignment, rate limit bypass, GraphQL attacks, and WebSocket hijacking.
   • Tested frontend vulnerabilities — XSS, malicious wallet injection, clickjacking, and missing SRI on CDN scripts.
   • Tested backend flaws — SSRF, signature replay, injection, deserialization, and SSTI/RCE.
   • Tested Web3-specific vectors — RPC key leakage from JS bundles, chain spoofing, ENS/oracle manipulation, and Subgraph abuse.
   • Tested infrastructure — secrets leakage, S3 misconfigurations, container security, and CI/CD pipeline vulnerabilities.
      
6. Automated Scanning
   • Ran Nmap, Nessus, Burp Suite Pro, OWASP ZAP, GraphQL Cop, Arjun, ScoutSuite, Prowler, and Trivy for comprehensive vulnerability and misconfiguration coverage.
      
7. Vulnerability Analysis & Exploitation
   • Deduplicated all findings, scored each using CVSS v3.1, developed proofs-of-concept, and classified them by severity with defined disclosure timelines.
      
8. Reporting & Remediation
   • Drafted a full technical report with per-vulnerability detail, reproduction steps, remediation guidance, and a prioritised fix roadmap.
   • Reviewed client fixes, re-executed PoCs, and classified each finding as Fixed, Partially Fixed, or Open.
      
9. Final Delivery
   • Delivered full technical report, executive summary, evidence archive, and Certificate of Audit — all transmitted encrypted via PGP.
   • Conducted a handover meeting with root cause walkthroughs and a 30-day post-engagement support window.
      

QuillAudits' Strategic Approach to Kite AI’s Security Audits

Our approach to auditing Kite AI’s combined a security-first mindset, comprehensive threat modelling, and rigorous testing methodologies. By leveraging both white-box and black-box testing techniques, we conducted an in-depth assessment of the system. Throughout the process, we maintained transparency and clear communication with the Kite AI team, ensuring a collaborative and thorough security review.
 

Comprehensive Audit Discoveries and Remediation Strategies

• High Severity Issues (6): Critical vulnerabilities with high risk of exploitation and impact.
• Medium Severity Issues (4): Moderate risks that could affect security if not addressed.
• Low Severity Issues (5): Minor issues with limited impact, but recommended for improvement.
• Informational Severity Issues (7): Observations and best-practice suggestions with minimal impact.

Here is a breakdown of the critical vulnerabilities in audit discoveries and remediation strategies:
 

Audit Findings

1. Keys Exposed in Client Bundle

Description:

Sensitive keys were exposed in the client-side JavaScript bundle due to the use of NEXT_PUBLIC_ environment variables, allowing attackers to extract them from the browser and perform unauthorized transactions.

Impact:

Complete compromise of smart account operations, unauthorized transaction signing, and potential theft of funds.
 

2. Unauthenticated Force Update Endpoint Allows Claim Status Manipulation

Description:

The /allocations/force-update-claimed endpoint lacked authentication and authorization, enabling any user to mark any wallet address as “claimed” without verifying ownership or actual claim activity.

Impact:

Denial of legitimate claims, database integrity issues, and potential financial loss for users.
 

3. OAuth State Parameter Not Validated

Description:

The OAuth implementation generated a state parameter for CSRF protection but failed to validate it during the callback, making the flow vulnerable to CSRF attacks and unauthorized account linking.

Impact:

Account takeover, unauthorized access, and potential theft of user rewards and data.
 

4. Missing Wallet Signature Authentication on Airdrop Endpoints

Description:

Airdrop endpoints did not require wallet signature verification, allowing users to check eligibility and potentially claim tokens without proving ownership of the wallet.

Impact:

Unauthorized claims, privacy breaches, and bypass of access controls.
 

5. Quiz Creation Denial of Service (EOA Address Hijacking)

Description:

The /v2/quiz/create endpoint accepted arbitrary wallet addresses, allowing attackers to create quizzes on behalf of other users and block them from earning rewards.

Impact:

Denial of service, disruption of reward mechanisms, and loss of user incentives
 

6. Clickjacking Can lead to theft of the user’s funds

Description:

The /claim page is vulnerable to clickjacking because the site currently allows third-party framing. An attacker can embed the claim page in an invisible iframe and overlay a deceptive UI that causes victims to click the real wallet Connect / Claim buttons inside the framed page. Because the claim flow includes wallet interactions that launch signature popups, a successful clickjacking chain could lead to token approvals or transfers that result in direct financial loss.

X-Frame-Options is legacy and not supported by the Content Security Policy (CSP) recommendation model. The modern and authoritative control is Content-Security-Policy: frame-ancestors. Use frame-ancestors as your primary defence; X-Frame-Options may still be added as a fallback for very old browsers but should not be relied on alone.

Impact:

The vulnerability allows an attacker to trick users into unintended wallet connections, and Users may claim tokens, thinking they're interacting with a different site, whereas Legitimate tokens may be claimed on behalf of users without their awareness and subsequently, theft of the tokens can be done by the attacker, leading to loss of funds to the user.
 

Remediation Strategies

1. Keys Exposed in Client Bundle

Recommendation: Remove NEXT_PUBLIC_ from sensitive variables, move signing logic to server-side, rotate exposed keys, and implement server-side cryptographic handling.
 

2. Unauthenticated Force Update Endpoint Allows Claim Status Manipulation

Recommendation: Add authentication (JWT), verify wallet ownership via signature, validate on-chain transactions, or remove the endpoint if unnecessary.
 

3. OAuth State Parameter Not Validated

Recommendation: Store state server-side, validate it on callback, use secure random generation, and implement expiration.
 

4. Missing Wallet Signature Authentication on Airdrop Endpoints

Recommendation: Implement EIP-191 signature verification, require signed messages with nonce and timestamp, and validate signatures server-side.
 

5. Quiz Creation Denial of Service (EOA Address Hijacking)

Recommendation: Validate user identity via JWT, bind wallet address to authenticated session, and enforce server-side checks instead of accepting user input.
 

6. Clickjacking Vulnerability Leading to Potential Fund Theft

Recommendation: Implement frame protection by adding Content-Security-Policy: frame-ancestors 'none' as the primary defense and X-Frame-Options: DENY as a fallback to prevent the application from being embedded in external iframes.
 

Conclusion

The audit uncovered critical weaknesses across authentication, wallet interactions, and claim logic—areas that sit directly on the path of user funds.

By addressing these issues, Kite AI eliminated multiple real-world attack vectors that could have led to unauthorized claims, account compromise, and fund loss. The fixes not only strengthened security but also restored the core guarantees of ownership and trust that Web3 systems rely on.

This engagement highlights a key reality: in Web3, even small gaps in validation or access control can turn into high-impact exploits. Proactively identifying and fixing these issues before production significantly reduced Kite AI’s risk surface and positioned the platform for a more secure launch.