Invisible, autonomous and hackable: The AI agent dilemma no one saw coming
1 min read
Summary
As generative AI becomes more popular, it raises several security questions for enterprises, including vulnerabilities and potential attacks, making multi-agent systems a difficult challenge for security experts.
Ideally, AI agents can be plugged into workflows to perform a variety of tasks, but this means giving them access to sensitive data, which must be done while ensuring that private information is not accidentally opened or sent to others.
This creates issues around accuracy and accountability, and it is unclear whether agents should be classified as digital employees or software, which impacts how much access they should have.
One answer to this issue could be to give agents specific identities in order to control access, while another is to use agentic platforms that allow users to observe the actions of agents, such as Pega’s AgentX.
Enterprises will have to rethink how they provide information access to users, which may lead to workflow overhauls.
While audits are not the perfect solution, more targeted answers may arise as companies adopt AI.