Protection Against Prompt Injection Attacks

Anthropic Launches Claude Science and Claude Sonnet 5

AI, Anthropic launches Claude Science AI research platform, Claude Sonnet 5 coding capabilities explained, Claude Science, Claude Sonnet 5, Anthropic, Artificial Intelligence
Facebook
X
LinkedIn
Reddit
WhatsApp
Source: JRdes / Shutterstock.com

Anthropic has launched Claude Science, an AI research environment designed for scientific workflows, and introduced the more cost-efficient Claude Sonnet 5 model with enhanced coding capabilities.

Anthropic introduced two new products on Tuesday. Claude Science provides a specialized workspace for scientific research, while Claude Sonnet 5 is a new AI model optimized for autonomous software development tasks and data analysis.

Ad

Database Integration and Computing Resources

Claude Science has been in development since October 2025 and runs as a desktop application directly on researchers’ devices rather than through a web browser. The platform combines multiple scientific tools and provides native access to more than 60 specialized databases.

The environment can visualize complex scientific data, including three-dimensional protein structures, genome browser tracks, and chemical structures. It also automates the distribution of computing workloads across local hardware and external server clusters. The tool is now available as a beta version for users on the Pro, Max, Team, and Enterprise plans.

Anthropic also announced its own preclinical research programs focused on developing medicines for neglected diseases. Eric Kauderer-Abrams, Head of Life Sciences at Anthropic, said: “These are areas that fall outside the scope of what the traditional pharmaceutical and biotech landscape might consider attractive targets, but they still represent a real burden.”

Ad

Sonnet 5 Capabilities and Pricing

The newly introduced Claude Sonnet 5 model was designed to independently plan and execute multi-step software projects. It can autonomously operate tools such as terminals and browsers.

Anthropic positions the model as a more affordable alternative to earlier versions such as Sonnet 4.6, while offering reasoning and programming performance comparable to the Opus 4.8 model.

For users, introductory pricing applies through August 31, 2026, at $2 per million input tokens and $10 per million output tokens. After that, pricing will increase to the standard rate of $3 and $15 per million tokens respectively.

Security Features and Export Restrictions Lifted

Sonnet 5 includes default filters designed to identify and block harmful requests. According to Anthropic, the model offers stronger protection against prompt injection attacks and produces fewer hallucinations than version 4.6.

During cybersecurity evaluations, the model underwent routine safety testing but was not capable of exploiting software vulnerabilities. Anthropic stated:

“Sonnet 5 was never capable of developing a fully functional exploit.”

Anthropic

Potentially dangerous use cases are intended to be blocked through real-time protection systems.

The launch coincides with a decision by the U.S. Department of Commerce, which lifted export restrictions on Anthropic’s Mythos 5 and Fable 5 models. The restrictions had been imposed in early June 2026 due to national security concerns.

(ll)

Ad

Artikel zu diesem Thema

Weitere Artikel