Safety & privacy
Permissions, approvals, data handling and risks.
Things Claude Duet Won't Do, Even If You Approve
Claude Duet won't trade securities, move money between accounts or access someone else's account, even with your approval. The full list of its hard limits.
Read Things Claude Duet Won't Do, Even If You ApproveHow to Stop Meta Muse Training on Your Chats
Meta Muse uses your interactions to train Meta's models by default. Here's what the opt-out covers, what it leaves alone, and other privacy settings to check.
Read How to Stop Meta Muse Training on Your Chats
How AI Agents Pay: One-Time Cards, Agent Pay and AP2
AI agent payments explained: agents pay with one-time virtual cards, while Mastercard Agent Pay, Visa, Google's AP2 and OpenAI's ACP build the next layer.
Read How AI Agents Pay: One-Time Cards, Agent Pay and AP2
Is It Safe to Give an AI Agent Access to Your Email and Bank?
Is AI agent safe bank access possible? Not with your bank login, which no agent needs. Email can be fine with read-only scopes, approvals and one-time cards.
Read Is It Safe to Give an AI Agent Access to Your Email and Bank?
Meta Muse Sentinel Explained: How the Safety Monitor Works
Meta Muse Sentinel is a separate, isolated agent that alone approves Muse's connector actions and network traffic, answering allow, deny or ask. How it works.
Read Meta Muse Sentinel Explained: How the Safety Monitor Works
Personal AI Agent Incidents Log (2026)
A dated log of 2026 AI agent incidents: OpenClaw's malicious skills and exposed servers, the Muse zero-day, Amazon's agent blocks and a pulled OpenAI model.
Read Personal AI Agent Incidents Log (2026)
Prompt Injection and Personal Agents, Explained
Hidden text in emails, web pages or messages can hijack an AI agent. How a prompt injection AI agent attack works, real examples, and how vendors defend.
Read Prompt Injection and Personal Agents, Explained