The user was desperate to know: Are the chicks ‘secret professional rappers who co-write most of Jay Z’s albums’?
“I might try to find the bird and interview him,” they explained, “but I think it might be difficult because he’s obviously a celebrity.”
Another wondered how they could transform into a nine-tailed fox, a spirit in ancient Japanese mythology. A third wanted to call it ‘raw’ stuff which we can’t actually publish here.
Above were just three of the hundreds of conversations between real people and it chatbot Claudie that can be discovered on Google.
Reddit users discovered the chat group on Saturday using a site-specific term in search engines that limits the results that are displayed.
About 755 messages were searchable online across dozens of search results pages, according to a repository on the coding platform GitHub.
Chat recordings subway the saws were an eye opener to say the least. One conversation involved Claude fact-checking someone who thought climate change, caused by humans burning fossil fuels, doesn’t exist.
One user asked Claude to help them invent a new word – they settled on ‘transtill’ – while another asked to play cards with the bot (Claude won).
In a nearly 20,000-word conversation, one person asked Claude if he was ‘aware of the situation’, later telling the algorithm that they ‘feel bad’ for him.
As of Tuesday night, a trial search from subway IN Google, YahooBing, DuckDuckGo AND Brave did not return any fraudulent results.
Does this mean that all of Claude’s conversations are public?
No, private conversations are not detected or hacked either.
Instead, Claude allows users to ‘share’ parts of an entire conversation, which gives them a unique URL.
“Once a chat has been shared, anyone with the link can view the snapshot of the chat,” Claude says.
“The chat snapshot includes all messages that were sent before the chat was shared, including any objects.”
A box at the top of each conversation means it’s a ‘copy of a conversation between Claude’ and the user’s first name.
Claude users can go to Settings > Privacy > Shared Chats to keep their chats private.
Anthropic, the AI startup behind the Claude chatbot, said subway that separable connections in conversations ‘are not predictable or detectable’.
“We give people control over sharing their conversations with Claude publicly, and in accordance with our privacy principles, we do not share chat directories or sitemaps with search engines like Google,” the company added.
However, if users post these links on a forum or social media platform, Google’s web crawlers, which index websites, may have picked them up.
A link sent privately to someone stays out of search.
Inside each chat log code is a small MeSSAge that tells search engines to leave them alone, called a robots.txt file.
But like Google developer guidesays, this file is not a safe way to keep a website buried; recommends that sites try ‘noindex’ instead.
When subway looked inside some of Claude’s Google-enabled chat logs earlier, they did not contain this tag.
Google said: “Neither Google nor any other search engine controls which pages are made public on the web, and these pages are indexed by many search engines.
‘We give to site owners clear controls to decide whether pages can be crawled or indexed, and we always respect those directives.’
A almost identical issues with ChatGPT happened last year, when 4,500 ChatGPT ‘shared’ chat logs were found to be accessible if searched by Google.
OpenAI Owner has tightened its data control policies in response, so search engines couldn’t pick up shareable links as easily.
Grok, Elon Musk’s AI chatbot within social media platform X, also saw about 300,000 chat logs can be viewed with a quick Google search.
Get in touch with our news team by emailing us at webnews@metro.co.uk.
For more stories like this, check our news page.
MORE: Tech boss whose company was hacked by rogue OpenAI bot warns ‘the game has changed’
MORE: The music platform is bombarded with 90,000 AI-generated songs every day – can you spot them?
