Synopsis: AI agents are racing ahead—and the guardrails are jogging to catch up. In this Techstrong.ai Leadership Insights conversation, Jitterbit CTO and SVP of engineering Manoj Chaudhary unpacks the security and governance gaps opening up as organizations sprint to put agents into production.
Architectures teams deployed six months ago already feel dated as new agent protocols flood in: Anthropic’s Model Context Protocol (MCP), Google’s agent-to-agent and payment protocols, IBM’s Agent Communication Protocol and more, all in under a year. Everyone wants an “AI agent” on the roadmap, but in the hype, security and governance are getting left behind.
Chaudhary draws a sharp line between innovation and risk. Autonomous agents without guardrails aren’t just prone to bias or hallucinations—they’re exposed to prompt injection, over-permissioned tools and data access that no one is systematically checking. Chaudhary argues that “human in the loop” shouldn’t be optional; validating agent output is going to become a profession in its own right.
Standards aren’t much of a safety net yet, either. Chaudhary points to ISO 42001 as one of the only AI-specific frameworks on the scene, focused on transparency, data quality, ethics and trust. At the same time, popular protocols like MCP still lack clear patterns for authentication, sandboxing, tool security and prompt-injection defense. Meanwhile, everyday behaviors—like dropping sensitive documents into public ChatGPT without opting out of training—are already leading to incidents, as Samsung learned the hard way.
Chaudhary’s prescription: treat AI accountability as a first-class design requirement. Track every input and output to and from LLMs and agents against guardrails, build cross-functional teams that include security and GRC from the start, and train design and dev teams on AI controls instead of bolting them on later.
Looking ahead, he expects AI agents to evolve into full digital identities—a “workforce” of autonomous agents, with other agents monitoring them for security and policy violations, and auditors and regulators not far behind.
