Muse Ember in Agentic Pentesting: Early Access Results

Arnica reports that, in early-access testing of Meta’s lightweight Muse Ember model within its agentic pentesting system, Ember outperformed a previous frontier-model baseline on speed, vulnerability recall, and reasoning-token use across five purpose-built applications. On its largest benchmark application, Arnica says Ember identified 80.7% of planted vulnerabilities compared with 61.4% for the comparison model, while across all applications it completed autonomous pentests in an average of 15 minutes rather than 35 and used 74% fewer reasoning tokens. The company describes Ember as a potentially cost-effective option for repetitive agent tasks and is considering making it the default model for Arnica Agentic Pentesting. However, its tests also found weaker recall on two smaller applications, occasional premature task completion, lower precision caused by reporting low-impact hardening issues as findings, inconsistent benefits from higher effort settings, and four failed calls in more than 24,000 requests.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.