Network Traffic Analysis of Freedom GPT: The Uncensored AI
FreedomGPT is a privacy-focused AI chat platform that promises “private access to 50+ of the top AI models” with user-driven leaderboard and subscription plans. Designed to provide uncensored, locally runnable, and privacy-preserving conversational AI experiences. Unlike mainstream LLM services that emphasize content moderation, FreedomGPT positions itself as a free speech–oriented alternative, allowing users to interact with AI models without guardrails or filtering. The platform supports a variety of large language models, including proprietary ones like Liberty, as well as other models such as Grok, OpenAI’s ChatGPT, Claude, LLaMA, Mistral, and Code LLaMA depending on deployment context (local, hosted, or hybrid). FreedomGPT offers a browser-based interface and also supports downloadable desktop apps for offline use. In addition to standard text chat, it provides features like ActorGPT, Voice Mode, and custom tool integrations, appealing to both end-users and developers interested in unmoderated AI agents.
Network Traffic Analysis
We have performed extensive user interactions with the FreedomGPT web application for approximately 7 to 10 minutes in a single session. The actions included submitting multiple text prompts to the chatbot, switching between available models, accessing the model selection menu, and updating user setting like recent model preferences.
Overall Analysis
The captured traffic was completely TLS 1.3 encrypted. We have further analysed the traffic based on host names.
Figure 1: Request-Response count per host
In the figure above we observe the maximum number of request-responses was seen by chat.freedomgpt.ai followed by us.i.posthog.com
Figure 2: Cumulative payload per host
Analyzing Endpoints
By examining the HAR file, we gain a detailed view of the HTTP requests and responses between the client and FreedomGPT servers. This analysis focuses on critical endpoints and their roles in the platform's functionality.
User Interaction:
Figure 3: Interaction with the chatbot
- Endpoint: /api/v1/chat/completions
- Method: POST
- Purpose: It is the core API that powers conversation with the GPT chatbot.
- In the post data section, we see additional custom flags like “showTwo”, “above18”, “customPrompt” and “deviceType”.
- The model.id being set to “auto” suggest dynamic model selection
- Response:
- HTTP Status: 200 OK
- Data Flow: Chunks of assistant responses stream back via “data:” lines.
- Model used: “gemimi-1.5-pro” from Google
Models Selection:
Figure 4: GET request for selecting AI models
- Endpoint: /api/models
- Method: GET
- Purpose: Get all available models for chat use.
- Request Headers:
- Content-Type: application/json
- Response Status: 200
Preprocessing layer:
Figure 6: Post request for processing
- Endpoint: /api/v1/chat/router
- Method: POST
- Purpose: This route likely evaluates the user query (topic, content type, flags) and returns the metadata on how the backend should handle it.
- Request Headers:
- Content-Type: application/json
- Response Status: 200 OK
- Response Body: {“type”: “text”}. Text likely means it’s an internal label indicating the type of response expected.
Update:
Figure 7: Post request for user update
- Endpoint: /api/user/update
- Method: POST
- Purpose: This endpoint saves user-specific metadata like recent model selection.
- Request Body: {“recentModel”: “hermes-3-llama-3.1-70b”}
- Response Status: {“success”: true}
Performance and Latency
In the observed trace, the wait time was the dominant contributor to total latency, often extending to several milliseconds. This aligns with the typical LLM inference delay, especially when using large transformer models. All requests were transmitted over persistent keep-alive connections using HTTPS, specifically HTTP/3 (QUIC) in this case- as evidenced by the httpversion:h3 in the HAR. The Har file confirms that TLS1.3 was consistently used with no downgrade. No signs of TLS session renegotiation or anomalous latency spikes were present, suggesting a well-optimized transport layer.
NOTE: While Freedom GPT can be useful, it is a prohibited tool by many companies and government entities. Policy and technical systems must be in place to prevent usage, and it is vital to confirm this via tests using BreakingPoint. These tests help validate the security measures and help organizations prevent accidental or malicious use of the platform.
FreedomGPT Traffic Simulation in Keysight ATI
At Keysight Technologies Application and Threat Intelligence (ATI), since we always try to deliver the hot trending application, we have published the FreedomGPT application in ATI-2025-10 which simulates the HAR collected from the FreedomGPT web application as of May2025 including different user actions like performing text-based queries, uploading multimedia files, using the generate image feature to create custom visuals and refining search results. Here all the HTTP transactions are replayed in HTTP/2 over TLS1.2.
FreedomGPT application and its 4 new Superflows as shown below:
Leverage Subscription Service to Stay Ahead of Attacks
Keysight's Application and Threat Intelligence subscription provides daily malware and bi-weekly updates of the latest application protocols and vulnerabilities for use with Keysight test platforms. The ATI Research Centre continuously monitors threats as they appear in the wild. Customers of BreakingPoint now have access to attack campaigns for different advanced persistent threats, allowing BreakingPoint Customers to test their currently deployed security control's ability to detect or block such attacks.