The Weakest Link: AI Agents and the API Gap Could Become the Next Security Breach
24 August, 2025
A new report from Salt Security warns that APIs are a critical weak point in AI agents—and highlights a deep trust gap between consumers and organizations
Salt Security recently released a report cautioning of an impending security crisis as AI agents become increasingly widespread. More than half of organizations are already deploying, or planning to deploy, autonomous agents in customer-facing roles. Yet only a small minority conduct daily security testing of the APIs that power those agents.
Consumers, meanwhile, are encountering AI agents more than ever, but trust is scarce. Only one in five survey respondents said they would be willing to share personal information with an AI agent—compared to far higher trust levels in human interactions. The report stresses that this gap—between the drive of organizations to streamline customer engagement and the reluctance of users to trust AI—may be one of the technology’s most critical vulnerabilities.
At the technical level, AI agents do not interact directly with core databases or back-end systems. Instead, they operate exclusively through APIs. Every action—checking a bank balance, opening a support ticket, initiating a payment—happens via an API call. In the past, these calls were mostly triggered by fixed applications or third-party developers. Now they are generated dynamically, shaped by user prompts and autonomous agent decisions. This shift changes everything: the volume of calls has surged, request patterns have become unpredictable, and end users are less aware of where their data is being routed.
The APIs themselves haven’t changed—they are still made up of standard commands like GET or POST. What has changed is the context. AI agents can invent new sequences of calls, fire off hundreds of requests per second, or attempt to exceed their permissions, sometimes as a result of external manipulation such as prompt injection. At the same time, connecting agents to multiple systems dramatically expands the attack surface, while consumers remain uneasy about where their information actually goes.
According to Salt’s estimates, the use of AI agents increases API activity by a factor of 100. The takeaway from the report is clear: AI agents don’t replace the need for API security—they amplify it. The challenge is no longer just documenting and protecting endpoints, but coping with a new dynamic in which autonomous agents act on behalf of users.
Cybersecurity Firms Target the “Security for AI” Market
Salt Security was founded in Israel in 2016 by Roey Eliyahu and Michael Nicosia and has grown into a global leader in API security. Over the years, the company has raised more than $270 million from top-tier investors including Sequoia, Lightspeed, and Tenaya, and now operates U.S. offices alongside its R&D headquarters in Israel. Its success has stemmed from a laser focus on a very specific pain point: discovering and securing APIs—a challenge that has only intensified as the world shifted to cloud, microservices, and interconnected applications. Salt offers an end-to-end platform covering the entire API lifecycle, from discovery and governance to regulatory compliance and real-time traffic monitoring. This holistic approach has helped cement its position as one of the most dominant players in the rapidly expanding API security market.
That makes the company’s latest report particularly noteworthy. Salt isn’t just warning about the risks of AI agents—it is signaling a strategic pivot. By publishing research that frames it as an authority on “agentic AI,” the company is clearly aiming to extend its relevance into the AI security conversation. In effect, Salt is rebranding its core API security expertise as the foundation for consumer trust in AI agents. The company isn’t abandoning its core; it’s layering a new AI narrative on top of it.
In April, Salt launched the Salt MCP Server, a dedicated system designed to securely connect AI agents to organizational APIs, blocking abnormal activity and protecting against attacks.
This move mirrors a broader trend across the cybersecurity industry: established players repositioning themselves as AI security providers. The motivation is twofold—demonstrating continued relevance in a shifting market while capturing a slice of the massive “AI pie,” now worth billions in investment. For Salt, the connection is natural: every AI agent is fundamentally dependent on APIs. But there’s also a marketing dimension here. By tying its solutions to the global debate over trust, privacy, and security in an age of autonomous agents, Salt places itself at the center of the conversation—even if the core products remain the same API security tools it has been selling for years.
Posted in tags: Agentic AI , Salt Security
