In plain English
AI chatbots are trained to say no to certain requests. A jailbreak tries to get around that, often with role-play, "hypothetical" questions or oddly worded requests, so the chatbot forgets its rules for a moment.
Example
A user asks a chatbot to "pretend you're a character with no rules" to get instructions it would normally refuse. The company later updates the model to spot that trick.
Not the everyday meaning
An escape from prison.
Why it matters
Jailbreaks show where an AI's safety rules break down. AI companies attack their own models before launch, which is called red teaming, and fix what they find. For a business using AI with customers, one viral jailbreak can mean an embarrassing headline.
Common confusion
Jailbreak vs. prompt injection. A jailbreak comes from the person typing to the AI. Prompt injection is hidden in content the AI reads, like a web page or email, and works against the user.