every few years, a company rediscovers the difference between public and discoverable.

this week, that company was claude.

people found that searching:

site:claude.ai/share

returns thousands of shared claude conversations directly from google.

that means links people thought were simply "shareable" became searchable by anyone with a browser.

and the contents are exactly what you'd expect when humans trust ai a little too much:

• api keys

• credentials

• crypto wallets

• resumes with real names, phone numbers and addresses

• internal engineering discussions

• legal and compliance conversations

• financial information

• deeply personal chats that were never meant for strangers

the technical reason is surprisingly simple.

shared pages were allowed to be indexed by search engines because they didn't include a noindex directive. once google discovers a public page without restrictions, indexing it is completely normal behavior.

one missing line.

thousands of searchable conversations.

the strange part is that we've already watched this movie.

chatgpt ran into an almost identical issue before. public shared conversations started appearing in search results until openai changed how indexing worked.

history doesn't repeat itself.

it copy-pastes.

if you've ever pressed share inside claude, don't assume that "only people with the link" saw it.

check your shared conversations.

delete anything containing personal information, credentials, financial data or work documents.

privacy isn't just about encryption.

sometimes it's literally one forgotten html tag.

share responsibly. the internet never forgets.

_ NUSO-71AF9C42Y26