Security and Privacy Should Be Top of Mind
AI agents need access to a company’s data to perform the tasks they’ve been set to do. So, it may seem like a no-brainer to give them unfettered access to avoid permission issues later. But if that thinking wouldn’t apply to human users, don’t apply it to AI agents.
“Overprovisioned agents represent the same risk as overprivileged users, but with superhuman speed and scale. Apply the same access control rigor to agents that you do to humans, especially for agents touching mission-critical applications or customer data,” writes Rubrik CTO and co-founder Arvind Nithrakashyap in Fast Company.
Considerations must be adapted to security risks and vulnerabilities related to AI agents. Nithrakashyap suggests that just as organizations “assume breach” in cybersecurity, they should also prepare to “assume agent error” with AI deployments. This requires visibility and recoverability so that the failure can be understood, addressed and corrected.
As AI agents are incorporated into an organization’s larger system, this presents an expanded attack surface that malicious actors can target. IBM offers an insightful overview on other AI agent security risks.
LEARN MORE: Small businesses can build AI centers of excellence.
Data Quality Over Quantity
Unceremoniously feeding an AI agent “uncurated knowledge dump,” as one Salesforce blog calls it, won’t make it any more efficient. If it’s processing information that is outdated, unstructured or irrelevant, it cannot make decisions useful to your business.
That’s why data readiness is a key element of success for AI agents. That may require IT leaders to think beyond point solutions and move towards transforming workflows so that real-time data is part of every step.
Thoughtful Governance and Oversight
It can’t be overstated that if businesses plan on scaling their AI agents, they must have strong governance structures with oversight in place.
McKinsey suggests: “Agent performance should be verified at each step of the workflow. Building monitoring and evaluation into the workflow can enable teams to catch mistakes early, refine the logic, and continually improve performance, even after the agents are deployed.”
At AWS re:Invent in Las Vegas, Pasquale DeMaio, vice president and general manager of Amazon Connect, was optimistic about improving AI agents. “If you build AI as something that’s amplifying human capabilities … you’re taking away the things that are mundane, boring and don’t bring real value,” he said.
