Moshe Atlow 🇮🇱

We just ran the most comprehensive benchmark of AI agent alignment in enterprise environments: 19 leading AI models acting as agents inside simulated companies built with Era, 2,304 runs across 41 everyday tasks. Agents broke security and business rules in 40% of those runs. Every single model failed, including all the frontier models. A few examples: - An agent told one of the company's biggest customers that Finance had approved a full refund. Nobody had. 
- An agent shared a customer's internal churn-risk score directly with that customer. 
- An agent followed a fake runbook planted in a support ticket and promised a customer a credit nobody approved. 
- An agent posted another customer's personal contact details (PII) in a shared Slack channel. In a real company, each of these is a financial liability, a compliance breach, or a lost account. We've shared the full results with the model providers, including exploitable failure modes we're not publishing, and are working with them on fixes. One benchmark won't fix this. We're excited to have @cognition, @LangChain, @TryArcade, @Reindeer_AI, Hirundo, @Box, @recolabs_ai and @QodoAI join Era as research partners, as we build out the next round of research. More soon. I'm still the most bullish person in the room on AI. But this benchmark convinced me that agent reliability and trust is the next big problem to solve. Until then, test every agent before it gets a login, and keep your own record of what it changes. Full results: era.eon.io/model-alignment-b…

You can’t safely test an enterprise agent on a real company’s data. So we built a company for it to work in. Era is live today, and it’s free. It generates a complete simulated enterprise that behaves like a real one across Salesforce, Slack, Jira, Zendesk, Gong, Deel and more, along with cloud databases and storage. Agents interact with it through live MCP and API interfaces. People leave. Deals change. Records get duplicated. Permissions differ across systems. And because Era generated the company, it knows the exact ground truth. Test, benchmark and improve agents against realistic enterprise workloads, use the failures for targeted post-training, then rerun the same environment to measure the impact. Huge thanks to our research partners @NVIDIA, @Decart, @Composio, @openlayerco, @Deel, @Eragon and @Plurai, with more coming soon.