DEVCORE Warns LLMs Are Opening New Gaps in Corporate Web Security

Companies are rapidly connecting large language model (LLM) assistants to internal data, account permissions and web services, potentially turning prompt-injection attacks or logic flaws into unauthorized actions and sensitive-data leaks. DEVCORE said AI security is not merely a model-level issue but also involves existing website architecture and access controls, with implications for an organization’s overall cybersecurity risk.

At its 2026 conference and in a security alert issued on March 20, DEVCORE shared red-team findings involving logic bypasses, access-control gaps and data leaks. It also analyzed risks involving Wi-Fi chips and software supply chains. Experts said companies defending against emerging LLM attacks must continue patching traditional web vulnerabilities and enforce the principle of least privilege.