Security & AI news

WIRED’s guide to using AI chatbots privately · By FireAI Security & Research Team · Published

Promises, proxies, enclaves or your own Mac: the four levels of AI privacy, according to WIRED

WIRED ranks how private AI really is, from policy promises to hardware enclaves to models that never leave your computer. Where FireAI’s own AI sits, and why.

“If the tech industry sought to create a method of seducing users into sending their deepest, most sensitive secrets to a server in a faraway data center, it would be hard-pressed to create a better honeypot than an AI chatbot.” That’s how WIRED’s Andy Greenberg opens his guide to using AI without giving up your privacy. The guide is worth reading in full; here is its ladder of options, from weakest to strongest, and what it means on a Mac.

Start from zero

WIRED’s baseline for the big three chatbots (ChatGPT, Claude and Gemini) is blunt: expect “approximately zero real privacy” from anyone determined to get your conversation records who has a legal path to them. That includes the service itself, its partners, contractors who help fine-tune it, law enforcement, or someone who subpoenas the records in a civil lawsuit.

You have this intelligent thing staring back at you, and you’re basically telling it, one question at a time, every possible thing there is to know about your life.

Matt Green, Johns Hopkins University, quoted in WIRED

Level 1: contracts and promises

The strongest cloud option is a contract: “zero data retention” (ZDR), which OpenAI, Anthropic and Google offer for enterprise versions. It isn’t available to ordinary users, and even it has exceptions, WIRED reports. Consumer services instead offer promises not to log your chats. Proton’s Lumo, for example, relies on a no-logs policy rather than end-to-end encryption, and users have to trust that it keeps its word.

Level 2: proxies

Services such as Duck.ai and Venice.ai relay your requests to models like Claude or ChatGPT, limiting what the underlying model learns about you. But the messages themselves still reach that model, and as Green points out, even stripped of metadata they can identify you: ask about the best coffee shops in your neighbourhood and you’ve revealed where you live. “All that stuff tied together is basically a fingerprint for you,” he says.

Level 3: hardware enclaves

Better than promises are technical limits on what the operator can see. Some services run models inside a trusted execution environment (TEE), where queries are processed on isolated hardware and the server can prove the rest of the system can’t access them. WIRED describes Confer, the chatbot launched by Signal creator Moxie Marlinspike, as offering “perhaps the strongest consumer-level privacy protections currently available in an AI chatbot.” Meta uses a TEE for a more private Meta AI mode in WhatsApp, and Apple’s Private Cloud Compute has TEE-like properties for Apple Intelligence requests handled on its servers.

Level 4: nothing leaves your computer

Then there are models that run locally. Tools like Ollama, LMStudio and LocalAI guarantee a level of privacy that is “tough to obtain with any cloud-based AI tool,” WIRED notes, “by the simple fact that no data ever leaves your computer.” Apple, too, handles many simple requests with an on-device model. The trade-off is capability: Green warns that local models get stumped and hallucinate far more often than the big cloud models. “You can run these on any laptop,” he says. “They’re just dumb as rocks.”

Where FireAI sits, and why

FireAI made its choice at level 4. Its AI runs entirely on your Mac and never sends what it analyses to any server, ours included. That trade-off works because FireAI’s AI has one focused job, not an open-ended conversation: looking at a connection an app on your Mac is making and explaining, in plain language, what it probably is. A small local model is enough for that, and it means the details of what your apps do stay on your machine. See the on-device AI model.

And for the cloud AI apps you do use, FireAI gives you the layer WIRED’s guide can’t: control over the connections themselves. The world map shows where each app connects, per-app rules decide which apps may reach the internet at all, and Paranoid mode blocks known telemetry. None of that makes a cloud chatbot private; it makes sure nothing leaves your Mac that you didn’t allow.

As Green concludes in WIRED: “If you want real privacy, you might have to pay for it.” FireAI is paid once, 49 € and yours to keep, no subscription. Download FireAI and try it free for 17 days. FireAI is made by HisnLabs.

Sources