RELAYTO Penetration Test Report

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Engagement Information Overview Date : February 19, 2026 Target :  comp-ai.test.relayto.com Client : Relayto Assessment : Web Application Penetration Test - GrayBox Total Findings : 3 Critical : 0 High : 0 Medium : 0 Low : 0 Informational : 3 Overall Security Posture The security assessment identified one critical-severity finding, one low-severity finding, and three informational findings. While most components of the environment appear well-configured, the presence of a critical vulnerability required immediate attention. The previously identified critical finding involves unauthenticated API endpoints that expose sensitive user data, including personally identifiable information (PII) and administrative account details, and allow cross-tenant access. This issue could enable attackers to retrieve sensitive information without authentication, increasing the risk of data leakage, targeted attacks, and potential regulatory or reputational impact. The issue has been remediated, with corrective measures applied to the affected components . The low-severity finding has been fully remediated . While t he informational findings do not pose immediate risk but provide recommendations to further strengthen the overall security posture. Overall, the organization ’ s security posture has significantly improved following the remediation of the critical finding with only informational findings remaining.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Methodology Our penetration testing methodology is built on recognized industry standards, incorporating guidance from PTES, the OWASP Testing Guides, NIST 800- 115, PCI Penetration Testing Guidance, OSSTMM, and the MITRE ATT&CK framework. By aligning with these proven methodologies, we ensure a comprehensive and methodical evaluation of the target environment, whether it involves web applications, APIs, internal/external networks, mobile apps, AI/LLM systems, containers, hardware, or human-focused testing. We draw from several key security frameworks to deliver high- quality assessments: OWASP (Open Web Application Security Project): We apply the OWASP Testing Guide rigorously to identify weaknesses in web applications, with special focus on the OWASP Top 10, to strengthen your application security posture. PTES (Penetration Testing Execution Standard): PTES provides a structured methodology for network penetration testing from reconnaissance through post- exploitation to ensure a complete review of your environment. MITRE ATT&CK: By leveraging the ATT&CK matrix, we emulate real- world adversary tactics and techniques to uncover impactful, attack- relevant vulnerabilities. NIST 800- 115, PCI Penetration Testing Guidance, and OSSTMM: These frameworks help shape our overall testing strategy, ensuring consistency, robustness, and alignment with industry expectations.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Penetration Testing Process Engagement Preparation – Scope & Planning: We collaborate with stakeholders to define the scope, goals, rules of engagement, and operational requirements. Engagement Launch – Kickoff: Testing activities begin, supported by clear communication channels and coordination of resources. Information Gathering – Reconnaissance: We perform extensive passive and active recon to map the environment and identify initial points of interest. Vulnerability Assessment – Discovery: Automated and manual methods are used to identify weaknesses across all in-scope systems. Exploitation – Attack Simulation: We attempt to exploit identified vulnerabilities to validate their impact and assess the effectiveness of existing defenses. Post- Exploitation – Impact Evaluation (Optional): When required, we analyze the potential depth of compromise, including privilege escalation, lateral movement, and data exposure. Reporting – Findings & Recommendations: We deliver a detailed report with validated findings, impact analysis, and actionable remediation guidance. Retesting – Validation (Optional): After fixes are applied, we perform retesting to confirm that vulnerabilities have been effectively addressed. Engagement Closure – Debrief: We conclude with a formal debrief, presenting results and strategic recommendations to help strengthen long- term security resilience.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Risk Rating We assess each identified vulnerability using the Common Vulnerability Scoring System (CVSS), which evaluates the technical severity based on factors such as attack vector, attack complexity, required privileges, user interaction, scope, and impacts on confidentiality, integrity, and availability. The resulting CVSS score provides an initial measure of the potential technical risk associated with the vulnerability. Risk Rating Definitions: Based on the CVSS base score, we assign a qualitative risk rating to each finding: Critical : Vulnerabilities posing an extremely high threat to organizational data. These should be remediated immediately, as exploitation is often publicly accessible and may result in significant data loss or full server compromise. High : Vulnerabilities with a potentially severe impact on operations, assets, or individuals. Exploitation can lead to data breaches, escalated application access, or local shell access. Medium : Vulnerabilities that may have serious consequences when combined with other issues. They typically do not cause immediate system compromise but can serve as entry points in an attack chain. Low : Security issues that do not directly impact system functionality but could facilitate additional attacks or reveal useful information to attackers. Informational : Findings that do not constitute exploitable vulnerabilities but highlight missing best practices or default information exposure.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Scope and L imitation While the assessment was conducted with the highest level of diligence and adherence to industry best practices, the following limitations applied to the scope of this penetration test: Restricted Scope Boundaries : Only the assets, systems, IP ranges, applications, and functionalities explicitly listed in the agreed ‑ upon scope were tested. Any components not included in the defined scope were excluded from assessment. Time and Resource Constraints : The penetration test was performed within a fixed timeframe and with limited resources. As a result, not all potential attack paths or complex chained exploit scenarios may have been fully explored. Snapshot in Time : The results reflect the security posture of the environment at the time of testing. Changes to systems, configurations, deployments, or infrastructure after the engagement may introduce new vulnerabilities not captured in this report. Third-Party Components : External services, SaaS platforms, or hosted components managed by third parties were not tested unless specifically authorized and included in the scope.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Web Application Penetration Testing Findings Finding Title: Unauthenticated API Endpoints (Closed) Severity: (CVSS3.1): 3 .9 ( Low ) Description: The vulnerability occurs when API endpoints that expose sensitive application data are accessible without requiring authentication. This allows an attacker to retrieve sensitive information, including full user details, without possessing valid credentials. Exploitation requires only knowledge of the API endpoint structure, making exploitation easy. An attacker can exploit this by sending HTTP requests to the exposed endpoints using valid usernames, including those from other companies, to retrieve user records. This behavior indicates that the application does not enforce tenant isolation, allowing cross-tenant access to sensitive data. Successful exploitation could result in large-scale disclosure of personally identifiable information (PII), exposure of administrative account details and API keys, targeted attacks against high-privilege users across multiple organizations, and increased risk of regulatory, reputational, and compliance impact. Risk:  Unauthenticated access to PII and administrative account data .  Exposure of sensitive configuration details such as API keys .  Cross-tenant access enables attackers to retrieve data from other companies .  Targeted attacks against high-privilege users across multiple organizations .  Increased risk of phishing, credential stuffing, and social engineering .  Potential regulatory and compliance impact due to PII disclosure.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Evidence: Direct HTTP requests to the identified API endpoints returned successful responses without requiring authentication. Using usernames from other companies, the tester was able to retrieve full user records, demonstrating cross-tenant exposure. The responses included full user records, including administrator account details and associated sensitive fields. Figure 1 - Unauthenticated request to the /api/companies endpoint returning successful response with another company ’ s data. Figure 2 - Unauthenticated request to the /api/users/{username}/full endpoint returning successful response with sensitive data.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Figure 3 - Unauthenticated request to the same endpoint returning successful response with sensitive data of a user from another company. Figure 4 - Unauthenticated request to the /api/company/ endpoint returning successful response with company data.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Figure 5 - Unauthenticated request to the /api/companies/{company}/users/{username} endpoint returning successful response with company user ’ s data. Figure 6 - Unauthenticated request to the same endpoint returning successful response with another company ’ s user data.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Figure 7 - Unauthenticated request to the /api/users/comp-ai/ endpoint returning successful response with company user ’ s data. Remediation:  Enforce authentication on all user-related API endpoints .  Implement strict server-side authorization checks, including proper tenant validation for all user data access .  Remove sensitive fields such as API keys from user-facing API responses .  Apply the principle of least privilege to all exposed data structures .  Conduct a comprehensive review of all API endpoints to ensure consistent access control and tenant isolation enforcement . Affected Endpoints:  comp-ai.test.relayto.com/api/users/comp-ai/  comp-ai.test.relayto.com/api/company/  comp-ai.test.relayto.com/api/companies/  comp-ai.test.relayto.com/api/companies/{company}/users/{username}  comp-ai.test.relayto.com/api/users/{username}/full Vulnerability Remediation Status (02/13/2026): The vulnerability has been remediated across the majority of the previously affected endpoints. The remaining endpoints, however, remain vulnerable but do not expose sensitive information. For this reason, the issue is currently considered partially fixed and severity was downgraded to low.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Evidence: Figure 8 - Unauthenticated request to the /api/users/comp-ai/ endpoint returning successful response without company user ’ s data. Figure 9 - Unauthenticated request to the /api/company/ endpoint returning successful response with company data. Vulnerability Remediation Status (02/19/2026): The vulnerability has been remediated, and the issue is now considered fixed.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Finding Title: Missing Content Security Policy (CSP) (Closed) Severity: (CVSS3.1): 3.7 ( Low ) Description: The issue occurs when the application does not implement a Content Security Policy (CSP). This allows an attacker to inject malicious scripts or other content into web pages, potentially leading to cross-site scripting (XSS) attacks, data theft, or session hijacking. Exploitation requires the attacker to have the ability to convince a user to visit a vulnerable page, making exploitation moderately difficult . An attacker can exploit this by crafting malicious scripts that the browser executes due to the lack of CSP. Successful exploitation could result in sensitive information exposure, user session compromise, and reputational damage. Risk:  No restrictions on which scripts, styles, or resources can be loaded .  Increased risk of cross-site scripting (XSS) .  Potential data leakage via injected content .  Users may be exposed to phishing or malicious scripts .  Reduces defense-in-depth for client-side security .

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Evidence: Figure 1- Scan result confirmed the absence of Content Security Policy (CSP) header. Remediation:  Implement a strong Content Security Policy header .  Restrict sources for scripts, styles, images, and other content .  Use 'default-src', 'script-src', and 'style-src' directives at minimum .  Test CSP using tools or browser developer consoles to ensure effectiveness .  Regularly review and update the policy as the application evolves . Affected Endpoints:  comp-ai.test.relayto.com Vulnerability Remediation Status (02/13/2026): The issue has been remediated, with the CSP header implemented. Thus, it is now considered fixed.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Evidence: Figure 2 - CSP header included in the application response.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Finding Title: Wildcard SSL/TLS Certificate In Use Severity: (CVSS3.1): 0.0 (Informational) Description: The issue occurs when a server presents a certificate that is valid for all subdomains under a given domain. This allows an attacker to broaden the impact of a single certificate compromise, as any subdomain covered by the wildcard may be impersonated if the certificate or private key is leaked. Exploitation requires access to the wildcard certificate ’ s private key or the ability to trick the certificate owner into deploying the key on insecure systems, making exploitation moderately difficult. An attacker can exploit this by compromising one system that uses the wildcard certificate and leveraging the exposed key to impersonate additional services or subdomains. Successful exploitation could result in credential theft, user impersonation, man-in-the-middle attacks, or unauthorized access across multiple subservices that rely on the same wildcard certificate. Risk:  Increased attack surface due to shared trust across multiple subdomains.  Single point of failure if the certificate/private key is compromised.  Potential for largescale impersonation or MITM attacks.  Difficulty in enforcing least privilege for certificates across services. Evidence: Figure 1 - The server uses wildcard SSL/TLS certificate. Remediation:  Replace the wildcard certificate with individual certificates per subdomain . Affected Endpoints:  comp-ai.test.relayto.com

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Finding Title: Missing DNS Records Severity: (CVSS3.1): 0.0 (Informational) Description: The issue occurs when specific DNS records such as DMARC, DKIM, SPF, DNSSEC, or CAA are not configured for a domain. This allows an attacker to abuse email systems for spoofing, phishing, or unauthorized issuance of TLS certificates, and could compromise the integrity of DNS responses. Exploitation requires the attacker to have access to send emails to external recipients or attempt unauthorized certificate issuance and additional effort planning, creating, executing the malicious attempt, making exploitation relatively difficult. An attacker can exploit this by sending spoofed emails that appear to originate from the domain, intercepting or spoofing DNS responses, or requesting certificates from Certificate Authorities without authorization. Successful exploitation could result in brand reputation damage, phishing attacks, or trust issues with users and partners. Risk:  Spoofed emails can bypass recipient email filters, leading to phishing or social engineering attacks.  Lack of DNSSEC may allow DNS cache poisoning or spoofing.  Unauthorized certificate issuance could allow man-in-the-middle attacks.  Failure to authenticate email sources may reduce email deliverability.

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Evidence: Figure 1 - DNS tool scan result indicating the absence of the records. Remediation:  Publish a DMARC record to define email authentication policies.  Configure DKIM for all outbound email domains to ensure integrity .  Implement DNSSEC to protect DNS integrity and prevent spoofing .  Configure CAA records to restrict which Certificate Authorities can issue TLS certificates for the domain . Affected Endpoints:  comp-ai.test.relayto.com

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Finding Title: Concurrent Logins Allowed Severity: (CVSS3.1): 0.0 (Informational) Description: The vulnerability occurs when the application permits the same user account to maintain active sessions across multiple browsers or devices simultaneously. This allows an attacker who gains unauthorized access to valid credentials to sustain a persistent session without detection, even if the legitimate user is already logged in. Exploitation requires the attacker to be authenticated or to have access to valid user credentials, making exploitation difficult . An attacker can exploit this by logging in from a separate browser or device and maintaining an active parallel session. Successful exploitation could result in unauthorized account access, reduced session accountability, and difficulty detecting compromised accounts. Risk:  Increased risk of unnoticed unauthorized access.  Difficulty in identifying simultaneous suspicious sessions .  Higher impact if credentials are leaked, shared, or phished .  Reduced session integrity and accountability . Evidence: Testing confirmed that a user could log into the application from two separate browsers simultaneously. Both sessions remained active and functional without restrictions .

Bubba AI, Inc. 2261 Market Street, San Francisco, California, 94114 Remediation:  Implement controls to restrict users to a single active session per account.  Revoke existing sessions when a new session is created, or notify the user of concurrent login attempts.  Provide session management interfaces for users to view and terminate active sessions.  Log and alert on unusual or concurrent session activity to detect potential compromise. Affected Endpoints:  comp-ai.test.relayto.com