You shared a link. You trusted the UI. And now your password is on page 2 of Google.
It's 9 AM on a Tuesday. You're sipping coffee, scrolling through your search results. You type "site:perplexity.ai" out of curiosity. And there it is — your entire conversation about that startup pitch, your competitor analysis, your half-baked login credentials, sitting in the open like a forgotten diary on a park bench. No lock. No warning. Just pure, unfiltered spill.
This isn't a hypothetical. It's happening right now. In the last 48 hours, security researcher Prontos uncovered a massive privacy blindspot: shared links from Claude (Anthropic) and Perplexity AI are being indexed by search engines, exposing everything from internal company documents to personal credentials. Claude has since fixed the issue — but Perplexity's data remains live, accessible to anyone with a browser.
I've been in this industry long enough to know that speed often beats perfection. But this time, the speed of exposure has outstripped the speed of protection. Volatility isn't just market action — it's the sudden realization that your private thoughts are public property.
The Context: How We Got Here
Let's rewind. Both Claude and Perplexity offer a "share" function — a way to beam a conversation to a colleague, a friend, or a client. The link is meant to be a secret handshake: only those who have it can see it. But the internet doesn't work on handshakes. It works on crawlers.
The problem is mundane but devastating: neither service applied the noindex meta tag or configured their robots.txt to block search engines from indexing shared URLs. For Claude, that meant thousands of conversation pages were picked up by Google's bots. For Perplexity, the same. The result? A buffet of sensitive data served to the open web.
This isn't new. OpenAI faced the exact same issue in July 2025 — shared ChatGPT links were indexed, exposing user data. The industry has had the blueprint for failure for years. And yet, again, we're here. It's a story of collective amnesia mixed with product design laziness.
I remember the ICO mania of 2017 — white papers rushed out, tokens sold on the promise of a decentralized future. We learned that speed without security leaves a trail of wreckage. This is the same pattern, just dressed in AI hype. We're repeating the mistakes of the past, but this time the data is far more intimate.
The Core: What's Actually Spilling?
Let me be precise. Prontos documented that both Claude and Perplexity shared links were being indexed. Claude's fix has been implemented — search results are now removed. But Perplexity? Still wide open. As of the time of this writing, a site:perplexity.ai search yields multiple pages of shared conversation links, many of which contain:
- Credentials — login tokens, API keys, even plaintext passwords.
- Resumes and personal details — full names, addresses, phone numbers.
- Internal company discussions — strategic plans, financial data, employee communications.
These aren't just trivial chat logs. They are the raw material of business. The kind of data that, in the wrong hands, leads to corporate espionage, identity theft, or worse.
Now, the technical root is simple: the shared URLs are generated without randomization or authentication headers that would prevent search engines from treating them as public pages. It's a classic case of treating a shared link as a security token, when in reality a shared link is just a URL — and URLs are what search engines eat for breakfast.
The real scandal isn't the bug — it's that Perplexity hasn't fixed it. Two days after the report, the files remain online. The company's interface says "Anyone with the link can view," but that's a lie. Anyone on the internet can find the link. The disparity between user expectation and reality is a chasm.
I've sat through enough post-mortem calls to know that the hardest part is admitting fault. But the silence from Perplexity is deafening. In the world of crypto, we learned that transparency is the only way to retain trust after a hack. The same principle applies here. Perplexity's failure to act isn't just a technical failure — it's a failure of responsibility.
The Contrarian: The Real Problem Isn't `noindex` — It's a Culture of Oversharing
Here's the angle everyone is missing: the real culprit isn't the missing meta tag. It's the product philosophy that prioritizes shareability over privacy. Both Claude and Perplexity — and OpenAI before them — designed sharing as a frictionless experience. Click, copy, send. No warnings. No permission layers. No double-checks.
They built for virality, not for safety. And that's the deeper issue.
In the race to capture users, these platforms optimized for the average use case — sharing a funny chatbot interaction — and forgot that the power user would share sensitive work. The design assumes the user knows what they're doing. But in cybersecurity, we know that assumption is deadly. Most users don't understand that a URL is a broadcast, not a whisper.
This is where my own experience kicks in. During DeFi Summer, I watched yield farmers copy-paste wallet addresses into Telegram chats, thinking it was safe. It's the same cognitive bias: "If I can see it, only I can see it." The human brain doesn't intuitively grasp the internet's architecture. And product designers keep exploiting that blind spot for simplicity.
The fix isn't just adding noindex. It's redesigning sharing to be opt-in, not opt-out. It's adding a confirmation dialog: "This link will be visible to search engines. Are you sure?" It's making the default state "private" until the user explicitly chooses "public." And it's providing a dashboard where users can revoke shared links at any time.
Claude has done the band-aid fix. But neither company has addressed the root — the culture of oversharing baked into their product DNA. Until they do, this will happen again. Not because of a bug, but because of a mindset.
The Takeaway: Trust Is the Only Currency That Matters
So where do we go from here? The immediate future: watch Perplexity. If they don't issue a public statement within 72 hours, expect a class-action lawsuit. The pattern is predictable — it happened to OpenAI, and it will happen again. Regulators are watching. The GDPR has teeth, and this is a clear violation of data minimization and consent principles.
For users: be paranoid. Don't share anything via an AI link that you wouldn't post on Twitter. The technology isn't safe until proven otherwise. For developers: stop treating shared links as a feature — treat them as a vulnerability.
I've seen the sprint, I've survived the trap. The companies that learn from this will be the ones that survive the trust reckoning coming to AI. The ones that don't will be left with a lot of hype and no one to share it with.
Are you sure you want to click that link? Maybe ask your AI first.