What are the real risks?
- Data sent to the model. When you ask about a page, its text goes to an AI provider. Check whether chats are stored, used for training, or used for ads.
- Prompt injection. A page can hide text written for the AI, such as an instruction to fill a form with your details. This matters most for agents that act on their own.
- Spending and sending. An agent logged in as you can buy or email as you, and some actions cannot be undone.
- Confident mistakes. AI can state wrong things fluently.
- Fake downloads. Popular new browsers attract lookalike installers.
How do I use an AI browser safely?
- Download only from the maker's own site.
- Read what the privacy policy says about chats and training.
- Use a private or incognito chat mode for anything sensitive.
- Before letting an agent act, ask: does it confirm before paying, sending or deleting, and can I stop it mid-task?
- Check important facts against the original source.
More on the agent part is in what an agentic browser is, and on chats in whether to paste personal info into AI.
How does Ace handle safety?
Ace is built on Chromium and distributed only from ace.ai. It blocks ads and trackers by default, does not sell your data, and does not use your conversations for targeted ads. Incognito Chats are never saved and never used for memory or model training. The assistant runs only when you ask, and it acts on the page in front of you: it does not go to other sites to buy or book unattended, so the clicks that cost money stay yours. The full list of what Ace collects and what the installer does is on the safety page.