Artificial Intelligence
In the era of the internet the a lot of problems arises. I think that there are no issue as big as AI and AI browsers. The problem is mainly the vast amounts of data that the servers has. We can do better than this, we can use local AI and anonymised AI. I never recommend to pay for AI as there are free ways to do it. So long as you are not connecting to a server and not allowing the AI to do things on your device, it isn’t dangerous. But it is dangerous if you let the AI take control over your device.
AI Models that you use is most likely an LLM which means “Large Language Model”.
Large language models have transformed the landscape of artificial intelligence, capable of generating human-like text and engaging in conversations. While they offer huge potential, there are a lot of concerns for both privacy and security we must achknowledge.
One major issue is that LLMs requires a fuck ton of data for training. We’re not only talking data also sensitive personal information which they collect from you as you write, generate pictures and whatever else you use the LLM for. The deployment of LLMs in applications can lead to situations where confidential information will be exposed.
The Usual Suspects
While ChatGPT, Claude, Mistral, LLaMA, Co-Pilot and so many other models can be useful we think that it is an emmense problem that all of these collect all the data that touches them.
The problem isn’t so much if you ask general questions like how to boil an egg or how many languages does the average European know. The problem is when you ask it for assistance and give it personal information.
Just take a second to think about the things you wrote to the AI. Then ask yourself if you would tell your neighbour this. If not, why would you tell Sam Altman, Mark Zuckerburg or Bill Gates these things?
The Solution
The solution to the problem is to use a local AI model that is disconnected from the internet. We will discuss this in the Local AI tab.
An AI browser uses artificial intelligence to enhance web browsing but introduces several concerns. The extensive data collection, allows for breaches. Data security is at risk as well when sensitive information is stored in an unsafe manner.
Privacy Risks
You give these AI browsers every data, you give them your name, birthday and passport so it can order tickets, you give it your address and country so it can order from Amazon and you talk to the AI itself everyday. This is enough to make any privacy conscious person think twice before even trying an AI browser.
The problem with an AI browser is that it demands so much data to do anything, it really demands to know you. The problem with giving anything about yourself to any entity is that it makes identity theft and financial fraud possible.
Security Risks
Comet, ChatGPT Atlas, Edge Copilot, Dia, Genspark, Opera Neon, Fellou and Sigma these are all agentic AI browsers. The day after OpenAI released the Atlas browser, experts already found out how to exploit it. Namely they injected the AI with dangerous prompts.
When you allow an AI to have the amounts of information that an AI browser needs to work and allow it to assist you, then a hacker could send you an email, that email you open in your browser, the email has small white text that the AI reads. The AI reads it as a command and now it returns and email with all information that it gathered over time. This is the reality of an AI browser.
Links
There is also other concerns, just imagine any text field that the AI has access to, a link for example can be used to inject a prompt as well, it might look legit but the AI browser will read it and do stuff.
Socials
Another risk with an AI browser is the fact that social media has all kind of people and if someone and it doesn’t have to be your friend, so long as the AI browser can read it, then you can be prompt injected.
Prompt injection is dangerous and these are only a few examples of why its dangerous.
