PART 14 · BLIND SPOT · THE AI MADNESS · 🟡
Conversational AI
While we fight to encrypt our messages, hundreds of millions of people pour their most intimate thoughts into ChatGPT, Gemini or Copilot, on US servers. This is the blind spot of the whole story. Sheer madness.
What you type into a cloud AI is a written, timestamped, stored confession. Worse than a message: you spill your anxieties, your health, your finances, your work secrets, often more honestly than to a friend. And unlike an encrypted messenger, it’s all readable in plaintext on the server.
What the facts say (2025-2026)Consumer tiers train on your data by default. In January 2026, a US judge
ordered OpenAI to produce 20 million ChatGPT conversations as evidence: the
affected users were neither notified nor consulted. Deleting a chat
does not guarantee it’s gone. And in one leak, 47,000 exposed conversations
held emails, phone numbers and re-identifiable intimate details.
The counter: local AI
What it's for Software to run an AI model directly on your own computer.
Why it matters
Your questions never leave the device: no server, no account, no telemetry, and it works
offline. Quality depends on your hardware, but nothing leaks. Ollama and LM Studio
are the easiest; Jan and GPT4All aim for maximum privacy.
For whom & when
Anyone who uses AI for even slightly personal topics and doesn’t want to hand them to a US
server.
Install & use
For a gentle start, install LM Studio or Jan (a graphical interface, like a
normal app).
Download a model offered in the app (for example Llama or Mistral), matched to your machine’s
power.
Chat away: everything happens locally. Ollama is the command-line version for the more
technical.
What it's for An app that runs AI models directly on your phone, fully offline.
Why it matters
Your prompts never leave the phone: no account, no cloud, no telemetry, and it works in airplane
mode. Open source (MIT), it downloads GGUF models (Gemma, Qwen, Phi, Llama) from Hugging
Face and runs them on your device’s CPU, GPU or NPU. The mobile counterpart to Ollama and LM
Studio.
For whom & when
Anyone who reaches for ChatGPT on their phone for personal questions and doesn’t want to hand
them to a US server.
Install & use
Install PocketPal from the App Store or Google Play (or build it from source).
Open Models, pick one that fits your phone, and download it.
- Load it and chat: everything runs offline, on the device.
What it's for
Cloud AI access, but built for privacy, when a local AI isn’t possible.
Why it matters
Duck.ai (DuckDuckGo) gives anonymised access to several models. Lumo (Proton)
encrypts exchanges and doesn’t train on them. Far better than consumer ChatGPT, but it’s still a
remote server.
For whom & when
Those who want cloud convenience without feeding the giants. Keep non-sensitive topics there.
The simple ruleNever paste into a cloud AI what you wouldn’t tell a stranger who’s recording: identity, health,
passwords, work secrets, confessions. For anything sensitive, use a local AI.