Shared Claude conversation links started showing up in public Google search results, putting discussions online that a lot of users probably assumed would only be visible to people who had the direct link.

It first came to light in a Reddit post on r/ClaudeAI. Someone shared a screenshot of Google results for the search “site:claude.ai/share,” and it turned up a bunch of clickable links to actual Claude conversation pages. Claude’s sharing feature lets you create public links so others can see the chat, handy for things like brainstorming or design work. The idea is that anyone with the URL can open it.

People familiar with the feature say the shared pages were supposed to have a “noindex” tag. That’s a standard web instruction that lets search engines look at a page but tells them not to list it in public results. Google and Bing normally respect it. In this case, some of the shared pages apparently didn’t have the proper noindex marking when they were crawled, so they got indexed and became findable through search. Even with a noindex tag, the pages would still work if you had the direct URL. The real problem was that those URLs became discoverable through Google.

The conversations that got exposed covered a pretty wide range of sensitive stuff: legal strategy talks, technical troubleshooting, personal chats, and other private or confidential material. Once they were indexed, anyone who ran the right search could see them.

Fortunately, the related chat logs have been removed from Google’s index. Searching “site:claude.ai/share” no longer brings up those links. If you’ve shared any Claude conversations, it’s a good idea to go check and manage them under Privacy > Shared Chats in the settings.

This kind of thing keeps coming up as more people use AI tools for work and personal stuff. Features that are meant to make collaboration easy can accidentally open the door to wider exposure when the technical safeguards (like proper indexing controls) aren’t fully in place or applied consistently. Other AI platforms have had similar slip-ups before.

Don’t miss a thing! Join our Telegram community for instant updates and grab our free daily newsletter for the best tech stories!

For more daily updates, please visit our News Section.

(Source: Reddit | ITHome)

The post Claude AI’s chat sharing links got indexed by Google, exposing personal chats appeared first on Gizmochina.

In a pretty big surprise for the AI world, China’s Zhipu AI’s latest model, GLM-5.2, just grabbed the top spot in one of the most respected creative coding benchmarks out there.

On June 19, 2026, Design Arena announced on X that GLM-5.2 has taken the #1 position in their single-round HTML web design leaderboard (non-agent category). The platform, basically the world’s first big crowdsourced benchmark where real people blindly vote on AI-generated designs, says it beat out Claude Fable 5, along with the Opus 4.6 and 4.7 versions. That’s a solid five-place jump from its predecessor, GLM-5.1, landing it with an Elo score around 1360 in the Code Categories arena.

What makes this win stand out is how the model actually designs. It’s apparently excellent at clean layouts, pulling in images from CDNs, typography, visual hierarchy, and those nice subtle animations that make a site feel alive. It also plays really nicely with popular libraries like Chart.js and Three.js. According to the results, it boosted its win rate by about 6 percentage points this time around.

One interesting detail: GLM-5.2 leans heavily on Tailwind CSS (showing up in 91% of its designs) and Font Awesome (51%). By comparison, Fable 5 only used Tailwind in about 57% of its sessions, which might explain some of the practical gaps people noticed.

Beyond the design quality, the pricing is a breath of fresh air. GLM-5.2’s API runs around $1.40 per million input tokens and $4.40 for output, way more affordable than Fable 5’s $10/$50. Plus, it’s an open-weights model under the MIT license with a solid 1M-token context window, which means teams can run it locally and tackle bigger, longer projects without being locked into a proprietary service.

Design Arena’s approach gives these results real weight: millions of votes from actual creators focused on real aesthetics and usability, not just synthetic tests. This isn’t just a win for web design; it puts GLM-5.2 among the strongest open models across broader coding benchmarks too.

For the AI community, this feels like another sign that Chinese labs are moving fast in specialized creative areas. While Claude Fable 5 is a strong player, GLM-5.2’s mix of strong performance, low cost, and openness could make it a go-to choice for a lot of web development teams going forward.

Don’t miss a thing! Join our Telegram community for instant updates and grab our free daily newsletter for the best tech stories!

For more daily updates, please visit our News Section.

(Source: Design Arena)

The post China’s GLM-5.2 beats Claude Fable 5 in web design, claims No. 1 spot appeared first on Gizmochina.

Anthropic has revealed a striking experiment where AI systems worked together to build a complete C compiler almost entirely on their own. Led by researcher Nicholas Carlini, the project shows how far autonomous AI collaboration has progressed in real-world software development.

How the Experiment Worked

The compiler was developed over two weeks using 16 independent Claude Opus 4.6. Each AI agent ran inside its own Docker container, cloned the same Git repository, and worked without a central controller or human manager. Tasks were picked automatically, conflicts were resolved through Git, and code was pushed upstream without supervision. In total, the agents produced around 100,000 lines of Rust code across nearly 2,000 coding sessions, at an API cost of about $20,000.

What the Compiler Can Do

The result is a fully functional, open-source C compiler written from scratch. It can successfully compile the Linux 6.9 kernel for x86, ARM, and RISC-V architectures, and handle major open-source projects like PostgreSQL, SQLite, Redis, and FFmpeg. On the demanding GCC Torture Test Suite, it achieved a 99% pass rate. As a symbolic milestone, it even compiled and ran Doom, a long-standing benchmark for compiler capability.

Why This Matters

This project demonstrates that AI systems can now self-coordinate, manage large codebases, and deliver production-grade infrastructure software. While the compiler still has limitations and is not yet a full GCC replacement, the experiment marks a major step toward long-running, autonomous AI-driven software engineering.

Impact on the Future of Coding

This experiment signals a shift in how software may be built in the future. For developers, AI agents could handle repetitive tasks, large refactors, testing, and bug fixing, allowing humans to focus more on design, architecture, and problem-solving. At the same time, it raises new questions about code quality, trust, and verification. While human programmers are far from obsolete, their role may evolve from writing every line of code to guiding, reviewing, and validating increasingly autonomous AI-built systems.

Read More:

• Tesla is Shifting From EVs to Humanoid Robots: Why Elon Musk Is Discontinuing Model S & X? new
• OpenAI Is Training Robots in Secret, and It Starts With Toasters and Clothes
• Why 2026 Will Be the Year of Humanoid Robots, Counterpoint Report

(via)

The post No Humans, Just 16 Claude AI Agents Built a Fully Functional C Compiler, Shocking Developers appeared first on Gizmochina.