Anthropic launches a new IA model with restricted cybersecurity functions: Claude Opus 4.7 improves reasoning and coding, but prioritizes global security.
From New York, the firm Anthropic revealed this Thursday Claude Opus 4.7, an update that only two months after version 4.6 elevates capabilities in software engineering and multidisciplinary reasoning. This model positions itself as the most robust option for general users and companies, surpassing competitors like Gemini 3.1 Pro from Google in coding tasks, according to the official statement from the company.
Restricted cybersecurity functions as priority
Unlike Claude Mythos Preview, presented last week with limited access due to high risks, Claude Opus 4.7 incorporates restricted cybersecurity functions. It includes automatic systems that detect and block high-risk requests linked to cyberattacks, making it “less capable” in this area to avoid threats to critical infrastructure.
The role of the Glasswing Project
This advance is based on the Glasswing Project, inspired by the transparent-winged butterfly that camouflages. More than 40 entities, including Amazon, Apple, Google, Microsoft, Nvidia, and JPMorgan Chase, access these defenses. The project arises from concerns about autonomous IA in finance and key systems, generating meetings between politicians and Wall Street bankers.
Effort control and accessibility
Anthropic, founded by ex-OpenAI engineers, introduces the “xhigh” effort level, allowing developers to balance deep reasoning with low latency in complex problems. The price remains at 5 dollars per million input tokens and 25 for output, available on Amazon Bedrock, Google Cloud, and Microsoft Foundry.
Context in the regulatory landscape
The launch coincides with scrutiny from the Trump administration towards Anthropic. Cybersecurity experts highlight that these restricted cybersecurity functions respond to a global consensus: the European Union and the US have pushed regulations like the AI Act to limit high-risk models. 2025 statistics indicate that IA-driven cyberattacks grew by 40%, according to CrowdStrike reports, justifying these measures.
- Claude Opus 4.7 leads in coding benchmarks.
- Restricted cybersecurity functions protect against abuses.
- The Glasswing Project unites tech giants in shared defense.
In an ecosystem where IA accelerates innovations, Anthropic launches a new IA model with restricted cybersecurity functions balancing power and responsibility. For the Dominican diaspora interested in technology, this step marks a milestone in the safe use of advanced tools, with implications for local developers in coding and digital security.
