Major security breach at Austrian AI startup localmind.ai (localmind.ai)

On 5 October 2025 Austrian AI startup Localmind disclosed a security breach that exposed parts of its infrastructure after a misconfigured external beta test instance granted a newly registered account administrator privileges. The attacker used that account to access an automation platform (Localmind Automate / n8n) where an overly-broad Notion API key was stored, allowing read access to the company’s entire internal knowledge base. That database contained infrastructure details and credentials, which the attacker leveraged to expand access and send spoofed e‑mails to customers. Localmind says the core production platform was not compromised; subsequent containment steps included immediate isolation of test systems, revocation and rotation of all passwords/API keys/tokens, mandatory 2FA for admins, deletion of exposed secrets, and notification to the Austrian data protection authority under GDPR. Forensics are ongoing and Localmind reports a follow‑up intrusion attempt on Oct 9 that was blocked, plus a brief unauthorized edit to an externally hosted webpage that has been corrected. For the AI/ML community this highlights perennial operational risks: unsecured test environments, poor secrets management, and excessive API key scopes can defeat otherwise robust products. Localmind is migrating VMs to Tier IV, ISO‑certified datacenters, performing per‑instance audits and manual hardening before restart, and offering customers secure data exports to assess exposure. The incident reinforces best practices—least privilege, isolated test networks, vaulted secrets, automation platform hardening and continuous monitoring—especially for startups packaging sensitive models and infrastructure.