Anthropic Releases Claude Opus 4.7 With Built-In Cybersecurity Restrictions
Anthropic unveils Claude Opus 4.7 as its most powerful generally available model, though it intentionally limits the AI's cyber capabilities compared to the restricted Claude Mythos Preview.
Anthropic releases Claude Opus 4.7 as its most powerful generally available artificial intelligence model, offering significant improvements in software engineering, instruction following, and real-world tasks over the previous Claude Opus 4.6. The new model outperforms its predecessor across various industry benchmarks, including agentic coding, multidisciplinary reasoning, and scaled tool use.
Despite these advancements, Claude Opus 4.7 is intentionally "less broadly capable" than Anthropic's Claude Mythos Preview, which the company currently limits to a select group of organizations through a cybersecurity initiative called Project Glasswing. Anthropic experiments with techniques to differentially reduce the cyber capabilities of Claude Opus 4.7 during its training phase.
The company builds automatic safeguards into Claude Opus 4.7 that detect and block requests involving prohibited or high-risk cybersecurity uses. Anthropic states that the data gathered from these real-world deployments helps the company safely work toward an eventual broad release of its more powerful Mythos-class models, maintaining its long-standing reputation for responsible AI development.