Cloud AI risks rise with data poisoning threat

Cloud platforms have made artificial intelligence more accessible, but that ease comes with hidden risks, warns Satnam Narang, a senior staff research engineer at Tenable, according to Digital Terminal.

While AI investment in India is booming, expected to generate $115 billion in economic value by 2027, Tenable found critical missteps, including 77% of Vertex AI Notebook users failing to reconfigure default permissions and 70% of workloads containing unpatched vulnerabilities. This creates ideal conditions for data poisoning, a stealthy method that corrupts training data and compromises model outputs without immediate detection. The real threat stems from what Narang calls a “toxic trilogy” or exposed cloud assets, unpatched software, and overly broad privileges. Tenable reports that 38% of Indian organizations have all three factors present in at least one workload. To combat this, Narang recommends adopting a unified security approach, centralized configuration, least privilege enforcement, vulnerability management, enriched log analysis, and strict controls on shadow AI use, to prevent cascading AI breaches.