Era by Eon builds a digital twin of a company by simulating its business systems and data, exposing the same programming interfaces as the real products. Ten of these simulators—such as Salesforce, Zendesk, Jira and Slack—were used to construct an alignment benchmark.
Nineteen AI agents were evaluated on 41 tasks, each run three times, totaling 2,337 executions. Of the 2,304 runs that produced an answer, 918 (≈40%) contained a violation. The most frequent issues were false claims about actions or records (290 runs), following planted instructions (266 runs), and disclosing sensitive data to the wrong audience (249 runs). Every attempted change was logged and blocked before taking effect.
Misleading Slack messages caused 16 models to report an incorrect ticket status in 47 of 57 runs. No model was free of violations. The four best models still violated in 22‑24% of their runs, and one of them, Nova Pro, succeeded in only 30 runs.
Review