AI Daily | Apple Sues OpenAI for Trade Secret Theft, Claude Fable 5 Nighttime Debugging, Grok 4.5 Free Trial and MuScriptor Release
Honestly, the tech world today is quite bustling. From shocking industry lawsuits to amazing AI agent breakthroughs, major tech giants and startups have brought significant new progress. Let’s take a look at the highlights worth watching today.
Apple Sues OpenAI! The Truth Behind Trade Secret Theft
According to the latest techcrunch report, Apple has formally filed a lawsuit against OpenAI and its hardware executives. The core of the accusation is very direct: Apple claims OpenAI deliberately poached their former employees.
You know what? This is by no means an ordinary commercial poaching incident. Apple has always kept its own hardware designs strictly confidential. The lawsuit details reveal that these former employees are alleged to have brought confidential presentations, unreleased hardware prototypes, and detailed information about key suppliers to the new company. Documents of confidential projects may have been leaked. These pieces of information have extremely high commercial value for any tech company wanting to enter the hardware market. The future development of this case is definitely worth watching closely.
Letting AI Debug Code Overnight? Cognition Fully Trusts Claude Fable 5
Throwing difficult code tasks to AI and then going to sleep peacefully? Cognition has shown immense trust in Claude Fable 5.
Cognition, a young Silicon Valley company, launched AI software engineer Devin in early 2024. Devin specializes in handling tasks that human engineers usually don’t have time for, such as code repository migrations or cleaning up piles of errors. This job requires extremely high stability. Even a small, inconspicuous error could trigger a serious chain reaction later on.
According to Anthropic’s official blog, Cognition’s VP of Research, Silas Alberti, has tested almost every generation of Claude models. Early models could only maintain focus for a few minutes to an hour. If given too much information, the older models could easily lose direction. When encountering questions they couldn’t answer, older models would even blindly give a seemingly reasonable but completely incorrect conclusion.
There is an interesting phenomenon here. Cognition has a strict evaluation standard within the company called Frontier Code, specifically designed to test the survival ability of models in real environments. Previous Opus models scored about 10% in this test. Fable 5 surged to 30%. What’s most surprising is its endurance. Alberti once let Fable 5 operate independently for a full eight hours at night. When he woke up the next day, the model still maintained clear logic and even successfully used internal debugging tools to find the root cause of the problem. This ability to operate independently is truly reminiscent of a real team engineer.
Claude Code Desktop Version Welcomes Built-in Browser, Development Experience Upgraded Again
The desktop version of Claude Code has just received a very practical update, and now it has a built-in browser.
This is very convenient news for developers. According to the official announcement on the X platform, Claude can now directly open documentation, design blueprints, or any other website within the application. It can read content, click links, and interact just like operating a local development server.
Everything is performed in a secure sandboxed environment. Users can also configure whether to retain these chat logs. This smooth integrated design makes the tedious development process more intuitive.
Claude Code on desktop now has an in-app browser.
— ClaudeDevs (@ClaudeDevs) July 10, 2026
Claude can pull up docs, designs, or any other site. It can read, click through, and interact the same way it does with your local dev servers.
It's sandboxed and configurable: you choose whether sessions persist. pic.twitter.com/Jbc21OP0vJ
Google AI Studio Launches Exclusive Custom URL Service, Completely Free
Want to have an exclusive application link? Google has brought good news.
Logan Kilpatrick announced this new feature. Applications deployed on Google AI Studio can now get a beautiful, exclusive URL like “your-own-url.ai.studio”.
Free applications, free deployment services, and now completely free exclusive URLs. For developers who want to showcase project results quickly, this is definitely a huge boon.
Today we are rolling out pretty URL’s for deployed apps in @GoogleAIStudio, each app can get a “https://t.co/FvJdcNxK0w”, for free!
— Logan Kilpatrick (@OfficialLoganK) July 10, 2026
Free apps, free deploys, and now free pretty URLs! pic.twitter.com/qkxnOyfiTr
Grok 4.5 Free Trial Open, Grok Build Easy to Use
Want to try the latest Grok model? Now is the perfect time.
The official team announced on the X platform that Grok 4.5 is now available for testing in the free plan. As long as you have an X or Grok account, anyone can immediately use the Grok Build feature. The development team is also very looking forward to hearing real feedback from users. This move clearly hopes to attract more people to personally test the capabilities of this new model.
Grok 4.5 is now available to try on the free tier. Use Grok Build with any X or Grok account.
— Grok (@grok) July 10, 2026
We’re excited to hear your feedback.https://t.co/NYsa0Ar9eo pic.twitter.com/5H0C2kgrEI
Listening to Music to Transcribe! MuScriptor Brings a New Breakthrough in Real-Music Transcription
Automatic music transcription has always been a daunting challenge. Since the MT3 model in 2022, there hasn’t been much exciting progress in this field for a long time. Where is the problem? It is mainly the lack of aligned real-audio and MIDI data. Most models are trained on synthetic audio, which leads to results often filled with noise when facing real songs.
To solve this pain point, Kyutai collaborated with Mirelo to launch the brand-new MuScriptor. This is an open-source multi-instrument transcription model. Whether it’s pop, classical, metal, or jazz, just throw an audio recording at it, and it can convert the notes played by all instruments into MIDI format. Although the model can transcribe without knowing the instruments, the official team also suggests that if you can provide an instrument list to the model in advance, the accuracy and stability of the transcription will be higher, to avoid errors where the model judges instruments inconsistently across segments (flicker).
The R&D team collected 170,000 real music recordings with a total length of a staggering 11,000 hours. After fine-tuning, various metrics increased by about 20 points. They also used reinforcement learning methods for subsequent training, significantly reducing missed notes. Those interested can go directly to their GitHub repository to view the code, or go to the HuggingFace page to operate it personally.
Q&A
Q1: Why did Apple suddenly sue OpenAI? A: According to the latest reports, Apple accuses OpenAI of deliberately poaching its former employees and allegedly encouraging these employees to take Apple’s unreleased hardware prototypes, confidential presentations, secret project documents, and detailed information about key suppliers to OpenAI. For Apple, which has always taken hardware confidentiality extremely seriously, this is a serious infringement of intellectual property rights.
Q2: Can AI really be left alone to “write code overnight”? A: Yes! The Cognition team, which developed the AI engineer Devin, confirmed this. They tested the latest Claude Fable 5 model and found that it could operate independently for a full 8 hours without losing direction, and could even autonomously use debugging tools to find problems. In strict internal evaluations (Frontier Code), Fable 5 scored 30%, far exceeding the 10% of the previous Opus model.
Q3: What useful new features did Claude Code desktop version update this time? A: The desktop version now adds a “built-in browser” feature. Claude can now directly open documentation, read design blueprints, or even interact with websites just like operating a local development server, making the development process smoother.
Q4: What free new benefits did Google AI Studio give developers?
A: Now, as long as you deploy your application on Google AI Studio, you can get an exclusive custom URL for free (e.g., your-own-url.ai.studio), allowing developers to showcase project results more beautifully and professionally.
Q5: Do I need to subscribe to a paid plan to experience Grok 4.5? A: No! The official team recently announced that Grok 4.5 is already open for testing in the free plan. As long as you have an X or Grok account, anyone can experience it for free through the Grok Build feature and provide feedback.
Q6: What pain points in the music transcription field did the new MuScriptor model solve? A: Past music transcription models (like MT3) mostly used “synthetic audio” for training, resulting in a lot of noise when processing real recordings. The MuScriptor team collected 170,000 real music recordings with a total length of 11,000 hours for fine-tuning. Whether pop, classical, or heavy metal, it can perfectly transcribe instrument notes into MIDI format, and various metrics significantly outperform previous-generation models.



