
Hey there! Did you know that AI agents now have more access and connections to enterprise systems than any other software out there? It’s true! This actually makes them a bigger target for cyber attacks than anything else security teams have had to deal with before. Crazy, right? Spiros Xanthos, the founder and CEO of Resolve AI, recently mentioned at a VentureBeat AI Impact Series event that if these AI agents are targeted, it could lead to a major data breach or something even worse.
What’s the deal with MCP?
So, there’s this thing called Model Context Protocol (MCP) that’s making integration between agents, tools, and data a whole lot easier for enterprises. But here’s the catch – MCP servers are “extremely permissive,” according to Jon Aniano, SVP of product and CRM applications at Zendesk. It’s a bit of a security risk because they don’t have as many controls in place as APIs do.
As we move towards a future where AI agents have their own identities and access, things are getting pretty complex. Xanthos pointed out that we’re lacking a framework for these autonomous agents. It’s up to us, as developers and users, to figure out how to restrict their actions and ensure they can be trusted.
Who’s responsible when AI messes up?
AI is becoming more involved in user interactions, especially in platforms like Zendesk. But things can get messy when humans are interacting with AI, and then that AI is interacting with other AI. Who’s to blame if something goes wrong? It’s a bit of a gray area right now.
Zendesk is playing it safe by setting strict access limits and closely monitoring what actions AI can take. But as demand for AI services grows, we need to establish clear standards for how these agents interact with each other and with users.
What’s next for AI security?
As AI agents become more trusted, they may eventually have more permissions than humans. But for now, many enterprises are hesitant to fully hand over control to AI. Resolve AI is taking steps towards giving agents more autonomy in certain safe scenarios, but there will always be risks involved.
So, what can security teams do right now? Well, they can start by using existing tools to control access and review actions taken by AI. It’s all about gradually expanding permissions and ensuring that human oversight is always in place.
