Using artificial intelligence in software development speeds up processes but also introduces new risks. Experts warn of a new class of threats tied to AI agents that can act on a user's behalf and interact with code, files, and outside systems.
The core issue is that AI no longer just handles text or code—it now works with entire infrastructures. This raises the risk of accidental system disruptions, configuration changes, or sensitive data leaking outside the organization. Communication channels between the AI and the developer also pose a risk, as they can be intercepted or exploited by attackers.
Research shows that a large share of data leaks already stems from employees feeding corporate data into public AI services. That can include code snippets, documents, commercial data, and internal policies. With AI agents, the stakes are higher because they don't just respond to queries—they take actions like accessing APIs, reading files, and interacting directly with systems.
Experts warn that if access rights are misconfigured or oversight is lacking, AI can inadvertently expose keys, tokens, or confidential data. There's also the threat of context poisoning, malicious instructions, and mistakes in agent-to-agent interactions.
Separately, AI is emerging as a new vulnerability layer at the intersection of development and cybersecurity. These are issues that don't fit traditional security models, such as context poisoning, unsafe queries, tool abuse, and overly permissive service accounts.
Still, experts see AI as a valuable tool for code analysis and vulnerability detection. It can spot common flaws like SQL injections and XSS faster, adding another layer of review.
But relying on AI alone for security decisions is a mistake. Models can err, overlook business logic, and foster a false sense of safety. Experts advise using AI as a support tool, with final checks left to cybersecurity professionals.