AI Developer Platform Breaches Fuel Stolen-Credential Resale and Bedrock Abuse

At least a dozen AI developer platforms reportedly suffered attacks over the past month, with attackers primarily seeking customer secrets and AI access tokens to monetize. Wiz traced enterprise credentials stolen in one vendor breach to two downstream Amazon Bedrock abuse cases involving approximately 2.5 million API calls and likely more than $100,000 in combined usage at list prices. Attackers validated keys, enumerated available models and quotas, and generated traffic consistent with resold coding-agent access; some stolen credentials were used within 30 minutes of reported exposure. Wiz linked access to stolen keys to resale operations associated with “momo” and Xingjiabi, but could not definitively connect those operations to the initial compromises or high-volume abuse. No single vulnerability or confirmed actor connects all the incidents.

The disclosures show varied intrusion paths and impacts. Railway reported unauthorized access to internal systems and exposure of some customers’ encrypted secrets, prompting credential rotation and temporary build shutdowns; its preliminary assessment said attackers sought AI provider keys for distillation attacks. GitBook revoked all personal API tokens after an attack it said used AI agents, with possible exposure of integration credentials and documentation visitors’ questions. Chainstack and WorkOS reported attackers using the open-source Artex AI penetration-testing framework: Chainstack identified potential exposure of account information, password hashes and node credentials, while WorkOS reported access to business contacts and customer communications through a compromised internal sales application. Organizations using affected platforms should follow vendor rotation guidance, inventory stored AI credentials, and monitor model usage, quotas and billing for unauthorized activity.