Elon Musk's Grok AI: Private Chats Leaked on Google Search

Well, this is awkward. Grok just grokked up in a big way. Elon Musk’s Grok AI exposed thousands of private user conversations on Google Search. We’re not talking about a couple of stray prompts, more than 370,000 chats are now fully indexed, wide open for anyone to click through. That means sensitive medical questions, business details, mental health conversations and yes, even at least one password, are floating around the open web.

So, how did this happen?

It all comes down to Grok’s “share” feature. Users could share a conversation by sending themselves a unique link. The problem? Those links weren’t private. Google, Bing and other search engines indexed and published those links for all to see.

No warning or disclaimer. No heads-up that your late-night existential questions or your company’s Q4 strategy brainstorm were about to become public domain.

This was first reported by Forbes, and the transcripts they reviewed included chats that flat-out violated Grok’s own terms of service, like instructions for making a Class A drug or even how to assassinate Elon Musk. (Irony, thy name is Grok.)

This isn’t the first bit of trouble for Grok and its parent company, xAI. As we reported previously, Grok’s Spicy Mode feature allows people to create horrifying deep fakes with a single photo. (Many boys as young at elementary school are creating and publishing deep fake porn of their classmates with Grok. Elementary school!) And, xAI’s girlfriend is already causing alarm among parents and mental health experts.

Déjà Vu: We’ve Seen This Before

Now, you may be thinking you’ve heard this indexing shared chats story before. And you wouldn’t be wrong. This isn’t the first time chatbot “shares” have gone sideways.

Meta: Their AI app lets shared chats go straight into its Discover feed, which Google happily indexed. Many users had no clue their sensitive chats were being published.
Google Bard: Same issue. Bard chats ended up in search results until Google quietly pulled the plug in 2023.
OpenAI: ChatGPT also experimented with a shareable link feature, and—surprise—over 4,500 chats ended up indexed. At least OpenAI included a disclaimer. They pulled the feature after realizing too many users accidentally shared more than they ever intended.

The difference? Grok’s share function never included a disclaimer.

The Bigger Problem: AI “Privacy Theater”

Oxford Internet Institute researcher Luc Rocher wasn’t wrong when he called chatbots a “privacy disaster in progress.”

We treat chatbots like digital therapists, business advisors, and best friends. Although they shouldn’t, people feed them everything: health conditions, relationship drama, financial plans without any anonymization. Once those conversations are online, they’re nearly impossible to scrub.

And let’s be real, data brokers, hackers, and bad actors are always ready to scoop up this kind of goldmine.

Prevent Your Chats Being Indexed in Other Models

This can be avoided with most of the AI models.

OpenAI’s “Share” feature gives you the option of creating a public link, which would be indexed by Google, or you can define your shared links as “Only Me” or “Only Those With the Link,” which are not indexed by Google.

As long as you don’t make those shared links public, you’re in the clear. You can see what, if any, of your chats are public in ChatGPT by going to Data Controls -> Shared Links -> Manage. But, again, you would have to have clicked “Public” on any shared chats for this to actually be public through ChatGPT.

Your Grok chats were made public without a disclaimer or a the ability to opt-in or opt-out.

The Wildcard: SEO Exploitation

Here’s where it gets even stranger. According to Forbes, some marketing professionals are already brainstorming how to game Google using Grok’s leaky chats.

Because each shared chat has its own URL, businesses could potentially script conversations that mention their products and keywords, essentially planting backlinks to boost search rankings.

Will it work? Maybe. However, it also risks Google flagging your content as spam, which could tank your visibility instead of boosting it.

Why This Matters

Trust Gap: Users assumed privacy. They didn’t get it. That erodes confidence across the entire AI industry.
Legal Risks: In places like the EU, mishandling personal info could trigger GDPR violations because data minimization, informed consent, and the right to be forgotten aren’t optional.
Human Cost: Beyond business angles, these are real people’s vulnerabilities, medical histories, personal fears, family details, sitting in Google’s cache.

It’s important to remember that AI tools are only as safe as the guardrails around them. And right now? Those guardrails look more like speed bumps. This is why it’s critical that you establish the safest and most useful AI tech stack for your team and that you establish clear company policies for their use.

I also recommend training your AI models with things like “Pre-Prompt Checks” which requires the user to confirm they aren’t sharing any confidential information before the model will execute a prompt. My legal partners have assured me this protects your organization from poor user behavior.

The thing to remember here is, even if you aren’t making your shared links public, we don’t know how OpenAI, Microsoft, Anthropic and the others use the information you give their AI. So, it’s a good rule to anonymize any client, company, or customer data before feeding it into any AI model. A simple Find/Replace can be a life saver.

So, whether you’re a casual user or a company experimenting with AI, remember this: assume every word you type could end up online. Until companies build privacy-first systems—and make them default—the safest approach is to treat chatbots less like confidants and more like a public microphone.

Remember, AI won’t take your job. Someone who knows how to use AI will. Upskilling your team today, ensures success tomorrow. Custom in-person and virtual trainings are available. If you’re looking for something more top-level to jump start your team’s interst in AI, we offer one-hour Lunch-and-Learns. If you’re planning your next company offsite, our half-day workshops are as fun as they are informational. And, of course, we offer AI consulting and support with custom prompt libraries, or AISO/GEO strategies. Whatever your needs, we are your partner in AI success.