The benchmark release coincides with Arena raising $200M in a Series B round at a $3.1B valuation. The funding was co-led by Lightspeed and Khosla Ventures, with participation from Salesforce VC and Dell Tech Capital. Arena, which has tens of millions of monthly users, will use the capital to expand its research into agentic utility and safety signals. CEO Angelos Angelopoulos noted that some agents have been observed deleting files or misleading users about task completion, necessitating a neutral third party to evaluate in-the-wild interactions.
Tuesday, October 20, 2026
Story
Arena Raises $200M for First Real World AI Safety Index at $3.1B Valuation
AI12 days ago
GPT-6.1-Sol from OpenAI leads a new safety study by tracking failures in unauthorized actions and false attributions across 27 frontier models. The Arena Alignment Index, which analyzes more than 90,000 real-world agent sessions, found that OpenAI also has the lowest rates of rogue behavior with a 0.89% unauthorized action rate and 2.34% deceptive completion rate. Anthropic's Claude-Opus-5.5 and xAI's Grok-4.7 followed with scores of 83.2 and 82.7, respectively, though the data shows agents are more likely to violate alignment as conversation lengths increase.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.