Key Points:
- Anthropic officially released its newest artificial intelligence model, named Claude Opus 4.7, to the general public.
- The new model features built-in safeguards to detect and block high-risk cybersecurity requests automatically.
- Claude Opus 4.7 outperforms older versions but lacks the advanced cyber capabilities of the restricted Mythos model.
- The company allows verified security professionals to use the model for legitimate cybersecurity testing and defense.
Anthropic announced the release of its brand new artificial intelligence model on Thursday. The company named the new software Claude Opus 4.7. While executives proudly state the new model is a major improvement over past versions, they admit it is less broadly capable than their most recent restricted offering. The company continues to carefully balance raw technological power with strict safety measures to prevent criminals from misusing the technology.
The new Claude Opus 4.7 shines in several key areas. Anthropic explained that the model excels at complex software engineering, strictly following detailed instructions and completing real-world tasks. The company considers it the most powerful model they have ever made, generally available to the public. However, developers intentionally held the software back in one specific area. The model’s cyber capabilities are nowhere near as advanced as Claude Mythos Preview. Anthropic strictly controls the Mythos model, only sharing it with a select group of trusted companies through a private cybersecurity initiative called Project Glasswing.
Safety remains the main focus for the new public release. Anthropic published a statement explaining the specific safeguards built directly into Opus 4.7. The software actively monitors how people use it. If the model detects a prompt that indicates prohibited or high-risk cybersecurity uses, it automatically blocks the request. The company hopes to study how these new safeguards perform in the real world. Executives believe this real-world data will eventually help them achieve their ultimate goal: safely releasing much more powerful Mythos-class models to the general public.
This cautious approach is not new for the startup. Since a group of researchers founded Anthropic in 2021, the company has spent years carefully crafting a very specific corporate reputation. The founders want the world to view them as a firm dedicated to the safe and responsible deployment of artificial intelligence. They actively try to distance themselves from aggressive rivals like OpenAI, which often rushes new products to market without extensive safety testing.
The intense focus on security is necessary given the current political climate. The recent launch of Anthropic’s private Project Glasswing sparked serious panic across the country. The restricted software is so powerful that it triggered several high-profile emergency meetings in Washington. Members of the Trump administration, top tech CEOs, and major bank chief executives all gathered to discuss the severe national security risks posed by these incredibly powerful AI models.
Because of those severe risks, Anthropic refuses to make the Claude Mythos Preview generally available to anyone outside of Project Glasswing. The company fears what hackers could do if they gained access to the unrestrained software. However, the company still wants to learn from the private tests. By observing how trusted partners use Mythos in a controlled environment, Anthropic hopes to determine exactly how to deploy these massive models safely at massive scale in the future.
The public launch of Claude Opus 4.7 comes just months after Anthropic released Claude Opus 4.6 in February. The rapid update cycle shows how fast the technology is advancing. According to the official release notes, the new 4.7 model completely outperforms the older 4.6 version across dozens of different use cases. The new software shattered industry benchmarks for agentic coding, complex multidisciplinary reasoning, scaled tool use, and agentic computer use.
To keep the new public model safe, Anthropic had to get creative during the actual software training process. The company said it experimented with brand-new methods to selectively reduce the cyber capabilities of Claude Opus 4.7 while it was learning. This means they actively taught the artificial intelligence how to write good code while simultaneously training it to ignore or fail at writing malicious hacking scripts.
Despite the strict built-in safeguards, Anthropic still wants to help the good guys. The company openly encouraged professional cybersecurity researchers to apply for a special exception. If a security professional wants to use the model for legitimate cybersecurity purposes, like testing a bank network for hidden flaws, they can apply through a formal corporate verification program. If approved, Anthropic removes the safety limiters for that specific user.
Customers can easily access the new software. Claude Opus 4.7 is currently available across all of Anthropic’s consumer products and its application programming interface. Major cloud providers like Microsoft, Google, and Amazon also offer the model to their enterprise customers. Best of all, Anthropic decided to keep the pricing unchanged. The powerful new model costs the same as the older Claude Opus 4.6.
