US startup Artemis Security has introduced Orion-1, an AI model built to autonomously investigate and respond to corporate network attacks. The model processes log data from cloud environments, authentication services, and network hardware to identify and link malicious activity. It assigns confidence scores to incidents and can initiate automated responses.
Artemis trained Orion-1 on its own security operations data, focusing on pattern recognition and detection rule creation. The company claims no customer data was utilized during the training process. The model is currently available in a private preview, allowing customers to set autonomy limits for specific actions, with high-impact tasks requiring human approval.
To evaluate performance, Artemis created the Decision-Grade Readiness benchmark. In testing across five tasks, Orion-1 achieved a score of 80.8 in attack reconstruction, outperforming the next-best model, which scored 58.3. The benchmark compared Orion-1 against models including Grok 4.7, Kimi K3, Claude Opus 5.5, and GPT-6 Sol.
Artemis Security exited stealth mode in April. The company has secured $70 million in funding through seed and Series A rounds, raised approximately six months after its founding. The development comes as general-purpose AI providers increasingly restrict cybersecurity capabilities in their models, often blocking security teams from analyzing autonomous attack data.
