What's trending in AI right now, from the app charts to the community feeds. Real links, our take.
Claude now marks what it writes
Starting August 2, all new Claude models embed a machine-readable watermark in their text output. The mark is invisible to readers but travels with the text when it is copied and pasted, so it can be detected later by anyone with the right tools. Anthropic rolled this out to comply with the EU AI Act, which now requires AI companies to label generated content. TechCrunch has the technical breakdown, and Forbes rounded up the reactions, which range from curious to alarmed.
The SEO world is particularly worked up about it. Search Engine Land ran an explainer this week for marketers who use Claude to generate content. The current read is that Google has not changed how it evaluates AI content, so rankings should not be affected right now. But the watermark does make that content easier to identify in the future if anything changes.
Grok 4.6 is out
xAI released Grok 4.6 on August 12, with a 500,000-token context window and a new high-effort reasoning mode aimed at longer agentic tasks. Third-party benchmarks put it level with GPT-5.6 Sol on the Artificial Analysis Intelligence Index, a step up from where Grok 4.5 landed. Pricing starts at $2 per million input tokens, roughly half what other frontier models have charged at launch. It is available through the xAI API and Cursor, with no open-weights release planned.
ChatGPT arrives on Linux
OpenAI shipped a native Linux desktop app for ChatGPT on August 11, bundling ChatGPT, ChatGPT Work, and Codex in a single package. TechCrunch confirmed the launch and noted that Anthropic had released a Claude Linux app about a month earlier. Phoronix has the package details: native .deb and .rpm builds for Ubuntu, Debian, and Fedora, on both x64 and ARM64. The app is in preview, so expect some rough edges, but if you have been running ChatGPT in a browser tab on Linux, there is now an actual desktop option.
AI short dramas keep climbing the entertainment charts
VibeShort, which delivers short AI-generated soap opera episodes in vertical video format, jumped six spots in the App Store Entertainment category this week. The App Store listing describes episodes of one to two minutes each, covering romance, fantasy, and revenge plots. The short-drama format took off in Asia a couple of years ago and is clearly finding an audience in Western app stores now. If you have been noticing clips from apps like this appearing in your social feeds and wondering where they come from, VibeShort is one of the bigger sources.
Running your own AI is gaining a real following
Two apps built for people who want AI without routing conversations through someone else's server had a strong week. Conduit, a native mobile front-end for Open WebUI that connects to your own self-hosted models, climbed 13 spots in the Productivity category. Private LLM, which runs language models entirely on your device with no internet needed after the initial model download, rose three spots. The common thread is privacy: both apps give you the benefits of AI tools without any of your conversations leaving your device. The category is still niche but has been drifting steadily upward.