How to Prioritise Your Privacy When Using LLMs
There’s nothing wrong with finding use cases for AI in your personal and academic lives, but there’s a real problem with AI assistants extracting more value from you than you might realise…
Know that Most LLMs store your conversations in plaintext — and can read them
With LLMs that weren’t built by privacy-first companies, it’s a fact of life that, until you delete a chat with an LLM (if you ever get round to it) then that information is being stored in plaintext on someone else’s server. This isn’t a private diary saved locally to your computer and protected behind your login credentials; it’s your information, thoughts, hopes, fears, and aspirations stored remotely somewhere in the world.
What does that mean in practical terms? It means that most mainstream chatbots retain those chats readably on their end for support, moderation, or product improvement, meaning a breach, a subpoena, or an employee with server access could expose it.
Understand that your career questions may be quietly training the next model
Do you know what you voluntarily opted into when you signed up for that LLM you’re currently telling about your weird dreams? In all likelihood, you’ll have to answer ‘no’ to that question, since more than 90% of people admit to not reading the terms and conditions.
Some AI assistants will retain and utilise chat data in order to improve or refine future responses, which means that chats that felt private at the time – and may be quite personal to you – could become training data for a company that didn’t see the value in prioritising your privacy.
Remember that free tools often monetize through data-sharing
Free services have to make money somehow, and AI companies have big overheads as far as data storage is concerned. The trouble is, if you’re not paying a subscription to use the service, one of the only options these companies have is to monetise your use of the tool…in other words, by passing your data onto third parties who see a lot of value in the insights it offers.
Targeted advertising has been a big business on the internet for many, many years now – long before the introduction of LLMs and AI assistants. For a long time, our data was totally out of our hands, being passed from one broker to another in order to extract as much value as possible out of our activities online. It’s why Europe introduced GDPR – to clamp down on the exploitation that was happening mostly out of our purview.
But GDPR and other data protection laws can’t offer much help if we (usually unknowingly) opt into agreements with LLMs designed to shuttle sensitive data from our keyboards to profit-seeking advertisers.
Stick with privacy-first companies
The internet is more like the wild west than ever before, however advanced it feels, but there are still places we can go to ensure we’re safe and that our best interests are being upheld. An AI assistant like Lumo, for instance, is built to offer all the benefits of an LLM without the major privacy risks and underhanded tactics deployed by so many of the other prominent players in the AI industry. Even Proton, the company behind it, can’t see your interactions with Lumo thanks to its encryption, and your data won’t be used to train future generations.
The only way to keep yourself truly safe is to stick to safe tools.