New research from Proton, the Swiss privacy company, reported by TechRadar, puts numbers on a contradiction: many people confide sensitive personal matters in AI chatbots while saying they do not trust the companies that run them.
Background
Proton commissioned the survey of UK AI users; it also runs its own privacy-focused chatbot, Lumo, and a tool called AI Paper Trail intended to show how much a typical conversation reveals. Proton is not a neutral party in a market it competes in, which is relevant to how the findings and quotes below should be read [1].
Findings
Two-thirds (66 per cent) of British AI users have discussed at least one sensitive topic with an AI assistant, according to the survey. The most common topics were personal finance (33 per cent), work or career problems (32 per cent), mental health (27 per cent) and relationships (20 per cent) [1].
At the same time, 42 per cent of UK users said they have little or no trust in AI companies to protect their private information, and 11 per cent said they don’t trust them at all. 64 per cent worry their conversations are being used to build advertising or marketing profiles, and 52 per cent worry their conversations are being used to train future AI models. 82 per cent said they would be more likely to use a chatbot designed around privacy [1].
The report’s own explanation for the contradiction is “freedom from judgment”: a chatbot answers at any hour and does not react with surprise or embarrassment [1].
AI is learning far more about us than a search engine ever could. We’re not just asking questions. We’re sharing deeply personal context about our health, finances and relationships. And when technology knows that much about us, privacy can’t be an afterthought.
Eamonn Maguire, Director of AI Engineering at Proton, quoted in TechRadar
Implications for Mac users
A privacy policy is a promise; where the computing happens is a fact. When a chatbot runs on a company’s servers, everything typed into it travels there, and what happens next depends on that company’s settings, its business model and whoever can legally demand its records. When a model runs on a person’s own device, what is given to it stays on that device. On-device models are smaller and, for open-ended conversation, less capable than large cloud models, but for a narrow job a small local model can be enough, and it is the only kind that cannot leak what it never sent.
Recommendations
- Open the chatbot’s settings and look for the option that lets conversations be used to improve or train its models; most services offer a way to turn it off, and it is rarely off by default.
- Look for a temporary or private chat mode for anything sensitive, and check how long the service says it keeps those chats.
- Leave out what identifies you or other people: full names, addresses, account numbers, health details, photos of documents. Describe a situation instead of pasting the record.
- Delete old conversations no longer needed; a history that isn’t kept can’t be exposed later.
- Be careful with chatbot plug-ins and connected accounts, since each one is a new place data can travel to.
Relevance to FireAI
FireAI’s AI, which explains in plain language what a connection on a Mac is, runs entirely on the Mac. What it analyses is never sent to a cloud AI and cannot end up in anyone’s training set. See the on-device AI model. For the AI apps a person does use, the world map shows where each one connects, and per-app rules allow deciding which of them may reach the internet at all. Paranoid mode also blocks known telemetry, so apps that report on a user in the background are refused.
FireAI cannot make a cloud chatbot private, and it cannot take back what has already been typed into one. Two habits help regardless of any firewall: check whether a chatbot lets its training use be turned off, and keep the most sensitive material out of cloud chats altogether.
Limitations
TechRadar’s report does not give the survey’s sample size, exact question wording or fieldwork dates beyond attributing it to Proton, and the survey covers UK AI users only, so the figures should not be read as representative of other countries. Proton’s commercial interest in a competing private chatbot is a reason to treat its framing, though not necessarily its numbers, with some caution.
Download FireAI and try it free for 17 days. FireAI is made by HisnLabs.