Anthropic has released a new artificial intelligence model, Claude Opus 4.7, describing it as its most capable publicly available system so far — while also emphasizing that it has intentionally reduced cybersecurity-related capabilities compared with its more restricted model, Claude Mythos Preview.
The company said Claude Opus 4.7 improves performance in areas such as software engineering, instruction-following, and complex task completion. However, it is still “less broadly capable” than Mythos Preview, which has only been given to a limited number of companies under a cybersecurity-focused program known as Project Glasswing.
Anthropic said it built safeguards into the new model that automatically detect and block potentially high-risk cybersecurity requests. The goal, according to the company, is to test these protections in real-world use before deciding whether to expand access to more advanced systems in the future.
The release reflects Anthropic’s ongoing positioning as a safety-focused AI developer, especially compared with rivals like OpenAI. It also comes at a time when governments and major tech firms are increasingly focused on the risks of powerful AI systems being used in cyber operations.
Despite the restrictions on certain capabilities, Claude Opus 4.7 is now available across Anthropic’s products, as well as through cloud platforms including Microsoft, Google, and Amazon.
Anthropic says it will continue refining its safety systems as it experiments with more advanced “Mythos-class” models, though there are no plans yet to release those widely.
Related Readings:
