Export
The link always opens the latest version of this page.
Ask an AI about it
Open in ChatGPTA new chat that reads this entry from the site Open in ClaudeThe same, in ClaudeShare the link instead of a file, and they will always see the latest version.
Files are a snapshot of today. The live page has anything newer.
ChatGPT Prompt Injection via Hidden Instructions
The attack leverages three primary delivery vectors 1:
- Pasted prompts — malicious instructions injected directly into conversation text
- Shared conversations — hidden directives embedded in previously saved or exported chat sessions
- Custom GPT builder instructions — adversarial system prompts injected during the creation of custom GPT configurations
The vulnerability allows attackers to override ChatGPT's intended behavior and safety guidelines by concealing instructions within seemingly benign user content or configuration parameters. This represents a class of large-language-model security risks where the model cannot reliably distinguish between legitimate user requests and adversarial instructions designed to manipulate its responses.
Facts 2
| Attribute | Value | Confidence | Sources | Cited |
|---|---|---|---|---|
| Date | 2026-06 | low | 1 | 1 |
| Impact | Hidden instructions in conversations could exfiltrate Gmail contents, chat history, and files to attacker-controlled accounts | low | 1 | 1 |