Last updated ·
Security & Trust
HackerAI is an AI agent for penetration testing and security work. This page describes the data we process, where agent code runs, and the services we rely on to operate.
On this page · 13 sections
Data we process
AI model providers
HackerAI sends your prompt, relevant conversation context, file content, images, and tool results to the company serving the model. Most model requests go through OpenRouter and an upstream model host. The host may be the model's developer, such as xAI, DeepSeek, Moonshot AI, Z.ai, or MiniMax, or another company serving the same model. See the OpenRouter privacy policy and upstream host data policies.
Some security-tuned models are served directly by Inference.net, and abliterated models are served directly by Abliteration.ai, without passing through OpenRouter. See the Inference.net privacy policy and Abliteration.ai privacy policy.
More than one provider can process a request. If a model fails or is unavailable, HackerAI may use a fallback on a different route; for example, Inference.net can fall back to Abliteration.ai. Uploaded images and browser screenshots may also be processed by a separate image model. The selected model does not identify every company that may receive relevant task content.
When the agent searches the web or opens a URL, Perplexity processes search queries and Jina AI retrieves page content. OpenAI screens message content for moderation.
Each provider processes data under its own terms. Retention and model training policies vary by provider, so we do not make a blanket no-training or zero-retention guarantee. Refer to the provider policies linked above.
Model training and data licensing
A future program may use eligible task content to train and evaluate AI models, including paid licensing to AI partners. This program is not active, and partner data collection has not started. See our Privacy Policy for the notice and consent requirements that would apply before any collection.
Execution environments
By default, terminal and browser actions run in an isolated E2B cloud sandbox. You can delete your sandboxes at any time from Settings → Data controls.
Local and Desktop modes run commands directly on your machine with your user's privileges and no E2B isolation, so we recommend them only on machines dedicated to testing. They are not offline: relevant commands, output, file content, and screenshots are still sent to HackerAI and the model providers above.
Account security
Where your data is processed
Your connection's IP address normally determines the initial region of a new cloud workspace: Europe uses our EU region (Frankfurt, eu-central-1), and other connections use a US region. Once a cloud workspace exists, Cloud Agent runs use its saved region even when your IP address changes. If you have existing workspaces in both the US and Europe, you may need to choose one or contact support.
New file uploads from the app normally follow your current connection location. Downloadable file copies uploaded by Agent runs normally follow that run's region. Uploads can use a US fallback location when regional placement is unavailable. Files that stay inside a Local or Desktop workspace remain on your connected computer. Changing your IP address through a VPN or travel does not move existing workspaces or stored files.
The primary database, AI model providers, web search, content moderation, authentication, billing, and analytics are not regional. See the Privacy Policy for international processing.
Billing
Data deletion
Subprocessors
Responsible disclosure
Incident communication
Source code
Compliance
HackerAI doesn't currently hold SOC 2, ISO 27001, or other third-party certifications. The service is offered in beta, as described in our Privacy Policy and Terms of Service. We'll update this page as our compliance program develops.