AI guardrails designed to stop hackers are now blocking defenders, and Orion-1 aims to fill the gap
decoder ·
Artemis Security has launched Orion-1, a specialized AI model designed to autonomously investigate and respond to cyberattacks on corporate networks. The system aggregates scattered log data from various sources to trace attacks back to single actors and initiate countermeasures. According to the company's proprietary 'Decision-Grade Readiness' benchmark, Orion-1 outperforms general-purpose models like Claude Opus 5.5 and GPT-6 Sol, particularly in attack reconstruction tasks. This launch addresses a growing issue where security guardrails intended to prevent hackers are inadvertently blocking legitimate defensive operations by major AI providers. While frontier models face increasing restrictions, Artemis offers a purpose-built solution fine-tuned specifically for security operations without using customer data. Orion-1 is currently available in private preview, allowing customers to set limits on autonomous actions with human approval required for high-impact decisions. The startup emerged from stealth recently, having raised $70 million in seed and Series A funding. Artemis claims its model bridges the gap between restrictive commercial AI tools and the needs of modern security teams.
decoder