[SECURITY]: Secure Publicly Exposed Actuator and AI Endpoints (Denial of Wallet Risk)

[pulkit1245]

Vulnerability Description

While reviewing SecurityConfig.java, I discovered an improper authorization vulnerability where sensitive and cost-incurring endpoints are exposed to the public internet without any authentication checks.

Specifically:

1. Spring Boot Actuator Exposed: The path /actuator/** is configured with .permitAll(). This exposes the application's internal management endpoints.
2. Unauthenticated AI Endpoints: Paths such as /ai/summarize, /ai/chat, and /ai/chat/ollama are also configured with .permitAll(). These endpoints interface with third-party LLM providers (e.g., OpenAI, Gemini), which charge per request/token.

I am participating in GSSoC '26 and would like to submit a Pull Request to fix this by restricting the /actuator/** endpoints to an ADMIN role and securing the /ai/** endpoints behind standard JWT authentication.

Impact

1. Information Disclosure (Actuator): By leaving the actuator endpoints unprotected, an attacker can access /actuator/env to read sensitive environment variables (such as API keys and database credentials) or /actuator/heapdump to download memory dumps containing sensitive user session data.
2. Denial of Wallet / Financial Loss (AI Endpoints): Because the AI endpoints are unauthenticated, a malicious actor can script a bot to send thousands of requests to the LLM APIs without needing a valid user account. This would quickly consume API quotas and result in massive, unexpected financial charges for the project maintainers.

Reproduction Steps

1. Start the backend application locally.
2. Open a terminal and run a simple curl command against the actuator environment endpoint without providing any Authorization headers:
   curl -X GET http://localhost:8080/actuator/env
3. Observe that the server returns the environment properties instead of a 401 Unauthorized response.
4. Similarly, make a POST request to an AI endpoint without a token:
   curl -X POST http://localhost:8080/ai/chat -H "Content-Type: application/json" -d '{"message": "Hello"}'
5. Observe that the request is processed and sent to the LLM provider instead of being blocked.

Severity Level

Critical

[pulkit1245]

Hi @viru0909-dev ,

I recently raised PR #1362 to address the critical security issue reported in Issue #1361 regarding publicly exposed Actuator and AI endpoints.

The PR secures:

• /actuator/** endpoints by restricting access appropriately.
• /ai/** endpoints by enforcing JWT-based authentication to prevent unauthorized usage and potential API cost abuse.

Since this is a security-related issue with potential information disclosure and denial-of-wallet risks, I would appreciate it if you could review it whenever you get time. If the implementation looks good, kindly consider merging the PR.

PR: #1362

Thank you!