DNey Guard: AI guardrails for ChatGPT and Claude

Checks your prompts on Claude or ChatGPT against your DNey guardrails. Blocks or redacts what breaks the rules.

What's new in 0.3.4 • On Claude (claude.ai): if one of your protections would mask part of a message, DNey Guard now stops the message and tells you which protection fired, so you can remove that part and send again. Before this version, the banner could say your message had been modified while Claude received the original text. • Fixed: on ChatGPT, while you were signed in, ChatGPT could send text you pasted to OpenAI before you pressed send, so before DNey Guard checked it. DNey Guard now blanks that early copy. • Fixed: on ChatGPT, DNey Guard could also check text from an editable draft in an earlier ChatGPT reply, as well as the message you sent. It now checks only your message. • Fixed: on ChatGPT, a message DNey Guard masked could be checked twice and show two banners. DNey Guard adds guardrails to ChatGPT (chatgpt.com) and Claude (claude.ai) in Chrome. When you send a message, it checks the text you typed or pasted. If one of your protections fires, it blocks the message or masks the matching part, for example a Social Security number. Every new account starts with three protections switched on: • Prompt injection: checks for attempts to override the AI's instructions. • Personal data: on ChatGPT, masks things like Social Security numbers, email addresses and phone numbers. On Claude, messages it would mask are blocked instead for now. Card numbers are blocked on both. • Data exfiltration: blocks links or images that would send your conversation to another site, and API keys and tokens, in what you type. You can change these, add others, and see a record of what was checked in your dashboard at dney.dev. What it doesn't do • It checks text only. It doesn't check the AI's replies, images, files you attach, or voice. • If Dney can't be reached, a personal install still sends your message and shows a warning banner so you know it wasn't checked. When an organization installs DNey Guard through Chrome's device policy, it blocks the message instead, unless the organization chooses otherwise. • No filter catches everything. Privacy To check a message, DNey Guard sends its text to the Dney service, which uses OpenAI to run some of its checks. Dney doesn't store what you write. It keeps a record of each decision: which protection acted, when, and counts such as message length. Your API key is kept only in this browser on this device. Privacy policy: https://dney.dev/privacy Free, no card. Set up takes under a minute.