Artemis Security launches Orion-1, a cybersecurity defense model

Artemis Security's Orion-1 processes log data from authentication services, cloud environments, enterprise endpoints, SaaS apps and network devices, linking isolated signals such as account logins, new cloud permissions and new mailbox rules back to a single attacker. The model attaches confidence scores to its conclusions and recommends or directly executes response actions. Artemis says general models lose the thread across billions of events or misclassify harmless but anomalous activity as an attack. Orion-1 was fine-tuned on the company's own security operations tasks, and no customer data was used in training. On Artemis's internal Decision-Grade Readiness benchmark, Orion-1 scored 80.8 on attack reconstruction, versus 58.3 for the best comparison model, GPT-6 Sol. The tests also included Grok 4.7 and Kimi K3, and only Orion-1 cleared the company's 80-point threshold on all five tasks. Orion-1 is in private preview for select customers on the Artemis platform, where customers can limit the model's autonomous decision scope by action type, with high-impact operations requiring human approval by default. Note that the comparison omits Anthropic's Claude Mythos 5 and OpenAI's GPT-6 Astra, and that Claude Opus 5.5 routes most cybersecurity tasks to the older Opus 4.8, so the benchmark results may mainly reflect Opus 4.8's performance.