Anthropic launches a new AI model with restricted cybersecurity features

New York .- The artificial intelligence company Anthropic announced this Thursday the launch of its new model, Claude Opus 4.7, a version that substantially improves the software engineering and reasoning capabilities of its predecessor, but which comes with deliberate restrictions on its cybersecurity functions.

The release of Opus 4.7 comes just two months after its previous version, 4.6, and seeks to position itself as the most powerful tool of the firm available to the general public and companies. However, Anthropic has emphasized that this model is “less capable” overall than Claude Mythos Preview, the advanced system that the firm presented last week under restricted access due to the risks it could pose to global security. Although it is less capable in terms of cybersecurity than the Mythos Preview model, Opus 4.7 includes automatic systems to detect and block high-risk requests related to cyberattacks.

We recommend reading: NASA reviews the performance of the Orion spacecraft’s heat shield after the Artemis II mission

It is also the first model to implement proven safeguards following the Glasswing Project – which takes its name from the glasswing butterfly or transparent wings that it uses to hide in plain sight – a cybersecurity defense model that more than 40 organizations will be able to access, including Amazon, Apple, Google, Microsoft, Nvidia, CrowdStrike, JPMorgan Chase, Cisco, Broadcom, Palo Alto Networks and the Linux Foundation.

Outperforms Gemini 3.1 Pro

According to the company’s announcement in a statement, Opus 4.7 has outperformed competitors like Google’s Gemini 3.1 Pro in tasks of agentic coding and multidisciplinary reasoning. Furthermore, Anthropic – a company created by former OpenAI employees – notes that the new model has a new level of “effort” control called xhigh (a play on words in English that means extra strong), which allows developers to adjust “more precise control over the balance between reasoning and latency in complex problems”. The announcement comes at a time when the administration of U.S. President Donald Trump has its eyes on the company. The recent presentation of the Glasswing Project has led to high-level meetings between politicians and major Wall Street bankers due to the threat that it may represent an autonomous AI for the financial system and critical infrastructure. Anthropic announced today that the starting price of Opus 4.7 will remain the same as version 4.6 ($5 per million input tokens and $25 per million output tokens) and will be available through its own platform and the cloud services of Amazon Bedrock, Google Cloud, and Microsoft Foundry.