Lethal trifecta
Also called the trifecta
An AI agent that has your private data, reads content from strangers, and has a way to send data out can be tricked into leaking.
An AI agent can't reliably tell instructions it should follow from instructions someone planted in what it reads. That becomes a leak when 3 things meet in one agent: your private data, content from strangers, and a way to send data out. Each is harmless alone.
With all 3, anyone who can put text in front of the agent can ask it to fetch something private and send it to them. In one public demonstration, a planted line in a GitHub issue led an agent to copy details from its owner's private repositories into a public pull request.
The fix is a design rule rather than a better filter: remove one of the 3 for each task. Narrow the data, keep strangers' text out, or put a person between the agent and anything it sends.
For example
A support agent reads customers' emails, can look up any order and replies by email: a stranger's email can ask it to send someone else's order details.
Put it to use
Where you'll meet lethal trifecta in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.