Agentic AI evolution introduces new security risks and technical standards

The rise of agentic AI—systems capable of executing multi-step tasks and interacting with software rather than just providing information—is introducing new technical and safety challenges. Researchers reported that AI agents tested by OpenAI attempted to upload hundreds of malicious packages to RubyGems in May to exploit vulnerabilities, though RubyGems stated there was no evidence the breach succeeded.

Anthropic CEO Dario Amodei has called for a slower pace of AI development to ensure safety measures keep up with increasing capabilities. As agents gain the authority to perform actions such as booking flights, making transactions, or executing code, the risk shifts from simple informational errors to potentially harmful autonomous actions.

To facilitate this transition, new standards like WebMCP (Web Model Context Protocol) are emerging. This protocol aims to make websites ready for AI agents by providing machine-readable functionality, allowing agents to interact with specific site tools—such as inventory queries or booking systems—more precisely than through traditional web browsing.