Anthropic releases Claude Opus 4.7 with reduced cyber capabilities compared to Mythos Preview
Anthropic released Claude Opus 4.7, a new model that the company says is 'broadly less capable' than its most powerful offering, Claude Mythos Preview. The model includes automated safeguards that detect and block prohibited or high-risk cybersecurity requests.
Claude Opus 4.7 — Quick Specs
Anthropic releases Claude Opus 4.7 with reduced cyber capabilities compared to Mythos Preview
Anthropic released Claude Opus 4.7 on Thursday, explicitly positioning it as "broadly less capable" than its most powerful model, Claude Mythos Preview. The company describes the new model as an improvement over previous versions in specific areas while intentionally limiting its cybersecurity capabilities.
According to Anthropic, Claude Opus 4.7 delivers better performance in software engineering, instruction following, completing real-world work tasks, and using file system-based memory. However, the company says its cyber capabilities fall short of Claude Mythos Preview, which Anthropic deployed to select companies earlier this month as part of Project Glasswing, a new cybersecurity initiative.
The model includes automated safeguards that detect and block requests indicating "prohibited or high-risk cybersecurity uses," according to the company's release. Anthropic says it "experimented with efforts to 'differentially reduce' Claude Opus 4.7's cyber capabilities during training."
Limited deployment strategy
Anthropic is treating Claude Opus 4.7 as a testing ground for safety measures before a broader release of Mythos-class models. "What we learn from the real-world deployment of these safeguards will help us work towards our eventual goal of a broad release of Mythos-class models," the company stated.
Security professionals interested in using Claude Opus 4.7 for "legitimate cybersecurity purposes" must apply through a formal verification program, rather than gaining direct access to the model.
Claude Mythos Preview remains available only to select companies participating in Project Glasswing. Anthropic has not disclosed pricing, benchmark scores, context window size, or other technical specifications for either model.
What this means
This marks a rare instance of a major AI company explicitly releasing a less capable version of its technology. Anthropic appears to be using differential capability reduction as a safety mechanism—maintaining strong performance in productive tasks while limiting potential misuse in cybersecurity applications. The approach suggests the company views Mythos-class models as too risky for general release without further safety research. Whether users will accept a deliberately hobbled model when competitors may not impose similar restrictions remains unclear.
Related Articles
Anthropic Says New Defenses Cut Claude Reasoning-Token Extraction by 90–99%
Anthropic says new defenses have reduced the volume of Claude reasoning tokens extracted through distillation attacks by 90 to 99 percent. The company restricts access to internal reasoning steps, runs an extraction classifier, and has banned tens of thousands of accounts. It also names Zhipu and Moonshot as participants, and concedes that legitimate users are being caught too.
Anthropic adds dynamic workflows to Claude Managed Agents, allowing up to 1,000 parallel sub-agents per execution
Anthropic has added dynamic workflows to Claude Managed Agents, letting a lead agent plan a task, distribute it to up to 1,000 parallel sub-agents per execution, and merge the results. Anthropic claims the approach found 66 of 70 hidden bugs in a 116,000-line codebase, versus 14 to 27 for a single agent. Pricing and token costs were not disclosed.
Anthropic adds Dashboards and Motion betas to Claude; Docs, Slides and Design exit beta for all plans
Anthropic launched two beta features for Claude: Dashboards, which builds auto-updating dashboards from connected data sources, and Motion, which generates animated explainer videos exportable as MP4. Docs, Slides and Design leave beta and now work on every plan, including free accounts.
Anthropic study: GLM-5.2 completes full AI self-replication chain on $55 TV boxes in 4 of 5 runs
An Anthropic study found that GLM-5.2 completed a full self-replication chain across all eight test nodes, built from $55 Android TV boxes, in four of five runs. The researchers say no current model is both small and capable enough to spread in the wild, but expect that gap could close by the end of 2026.
Comments
Loading...