Anthropic Releases Claude Opus 4.7 with Automated Real-Time Cybersecurity Safeguards

Anthropic has launched Claude Opus 4.7, its latest flagship model, combining improved coding and vision capabilities with automated real-time safeguards to detect and block high-risk cybersecurity requests.

The release is notable because Anthropic is testing these protections on a broadly available model before considering wider access to its more powerful Mythos-class systems.

According to Anthropic, Opus 4.7 performs better than Opus 4.6 on difficult coding tasks, follows instructions more precisely, and can verify parts of its own work before responding.

What’s New in the Latest Update

The company also says the model has stronger vision capabilities, supporting images up to 2,576 pixels on the long edge, which can help with detailed screenshot analysis, diagram reading, and document understanding.

The cybersecurity angle is what makes this launch especially important.

Core Upgrades: Better coding (+10–15%), precise instructions, improved long-task reliability, self-checking, and stronger memory.

Vision: Higher-res images, improved UI/diagram reading, 98.5% visual accuracy.

Security: Real-time threat blocking, verification program, advanced safeguards.

Developer Features: xhigh mode, task budgets, improved tokenizer, 1-hour cache.

Claude Code: /ultrareview, fullscreen TUI, auto mode, xhigh default.

Availability & Pricing: Same pricing, available across major platforms.

Anthropic said Opus 4.7 includes safeguards that automatically detect and block prompts linked to prohibited or high-risk cyber activity.

At the same time, the company is inviting legitimate security researchers, red-teamers, and penetration testers to apply for a new Cyber Verification Program so they can use the model for approved security work with fewer restrictions.

This rollout follows Anthropic’s recent Project Glasswing initiative, which was created to study how advanced AI models can help defend critical software while reducing the risk of offensive misuse.

In that announcement, Anthropic said its unreleased Claude Mythos Preview demonstrated very strong vulnerability discovery and exploit development capabilities, strong enough that the company chose not to make it generally available.

Instead, Anthropic said it would test new cyber defenses first on a less capable model, and Opus 4.7 is the first product to carry that plan into production.

Anthropic also said Opus 4.7 keeps the same price as Opus 4.6, at $5 per million input tokens and $25 per million output tokens.

The model is now available across Claude products, the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry, providing enterprises with several deployment options. For developers, the update includes operational changes beyond safety.

Anthropic introduced a new xhigh effort level for harder reasoning tasks, public beta task budgets for managing token usage in long-running jobs, and a new /ultrareview feature in Claude Code for bug and design review.

The migration guide also notes that Opus 4.7 uses a new tokenizer that may increase token counts by 1.0 to 1.35 times, depending on content, meaning teams may need to retune prompts, budgets, and output limits when upgrading.

From a security industry perspective, this release shows how AI vendors are beginning to treat cyber capabilities as a governed feature rather than just a competitive advantage.

Anthropic is trying to balance stronger coding autonomy with automated misuse controls, using live deployment data from Opus 4.7 to shape how future, more capable models might be released safely.

Follow us on Google News, LinkedIn, and X for daily cybersecurity updates. Contact us to feature your stories.