A real-world example shows an agent correctly denying a request for a specific company's data but leaking other firms' data on a generic prompt. This highlights that agent security isn't about blocking bad prompts, but about solving the deep, contextual authorization problem of who is using what agent to access what tool.

The ecosystem of downloadable "skills" for AI agents is a major security risk. A recent Cisco study found that many skills contain vulnerabilities or are pure malware, designed to trick users into giving the agent access to sensitive data and systems.

Unlike human attackers, AI can ingest a company's entire API surface to find and exploit combinations of access patterns that individual, siloed development teams would never notice. This makes it a powerful tool for discovering hidden security holes that arise from a lack of cross-team coordination.

Despite their sophistication, AI agents often read their core instructions from a simple, editable text file. This makes them the most privileged yet most vulnerable "user" on a system, as anyone who learns to manipulate that file can control the agent.

Developers are granting AI agents overly broad permissions by default to enable autonomous action. This repeats past software security mistakes on a new scale, making significant data breaches and accidental destruction of data inevitable without a "security by design" approach.

An AI agent capable of operating across all SaaS platforms holds the keys to the entire company's data. If this "super agent" is hacked, every piece of data could be leaked. The solution is to merge the agent's permissions with the human user's permissions, creating a limited and secure operational scope.

While sophisticated AI attacks are emerging, the vast majority of breaches will continue to exploit poor security fundamentals. Companies that haven't mastered basics like rotating static credentials are far more vulnerable. Focusing on core identity hygiene is the best way to future-proof against any attack, AI-driven or not.

Security researchers gained full read/write access to McKinsey's internal AI platform in just two hours via a prompt injection attack. This exposed 46.5 million confidential chats on strategy and M&A in plain text, highlighting severe security vulnerabilities in enterprise AI deployments.

AI agents are a security nightmare due to a "lethal trifecta" of vulnerabilities: 1) access to private user data, 2) exposure to untrusted content (like emails), and 3) the ability to execute actions. This combination creates a massive attack surface for prompt injections.