A startup called SplxAI Inc. is pushing for artificial intelligence agent developers to adopt a more offensive approach to security after closing on a $7 million seed funding round today.

The round was led by LAUNCHub Ventures and saw participation from Rain Capital, Inovo Runtime Ventures, DNV Ventures and South Central Ventures. Its backers believe there’s a growing opportunity for anyone who can help to secure AI agents, as these next-generation autonomous tools are expected to transform enterprise work in coming years.

AI agents are the next generation of generative AI chatbots like ChatGPT, capable of taking actions and performing other tasks autonomously on behalf of humans, with minimal supervision. They promise to transform the way work gets done, with one study showing that 33?% of enterprise applications will integrate with AI agents by 2028. But for now, the vast majority of these AI agents represent a critical security risk.

The challenge is the inherent vulnerability of the underlying large language models that power AI agents. Since the first generative AI chatbots emerged earlier this decade, hackers and other individuals have engaged in concerted efforts to manipulate the models through “prompt injection” attacks in an effort to trigger “hallucinations,” misbehavior or even try to compromise sensitive data.

AI developers have tried to prevent this through the use of guardrails and other protections. But such systems are often ineffective, and other security measures can be too restrictive, hurting the ability of the chatbots to work as they should.

SplxAI believes it can beef up agentic AI security by automating so-called “red teaming” efforts, which is where “white-hat” security professionals set out to try and uncover weaknesses in LLMs on behalf of developers. The startup attempts to simulate sophisticated adversarial scenarios that mimic prompt injection attacks. Its platform is built on a comprehensive database of attack methods that’s continually updated using a proprietary research engine, adding new vulnerabilities and threats as soon as they emerge.

Using SplxAI, organizations can proactively secure their AI agents by automating security testing processes and monitoring threat activity continuously via log analysis. That allows them to detect attacks as they happen and identify any vulnerabilities that are exposed so they can be fixed immediately, before an attacker exploits them.

The startup’s founding team has plenty of expertise in this area. Many of them are former AI red teamers with experience at companies such as Cisco Systems Inc., Zscaler Inc. and Wiz Inc.

The only real alternative to SplxAI is to conduct manual AI security testing or outsource this work to an external service provider, but in both cases, doing so is extremely expensive and also inefficient. The startup argues that manual risk evaluations are at least five times more expensive than its automated approach, and they take so long that they significantly slow down the rate at which organizations can deploy new AI agents.

SplxAI Chief Executive Kristian Kamber said the company is trying to redefine how AI practitioners stress test their agentic AI workflows. He insisted that only continuous and automated in-depth testing can uncover the most dangerous security threats.

“With the rapid advancement of LLMs, manual testing is not feasible,” he argued. “SplxAI’s advanced platform is the only scalable solution for securing agentic AI, providing security leaders with the tools they need to confidently embrace AI.”

Constellation Research Inc. analyst Holger Mueller said that as businesses race to create more powerful AI agents, they also need a better way to protect themselves against those agents being manipulated and used to hurt them.

“It’s good to see SplxAI bringing forward some protection against that kind of malicious activity, bolstered by a sizable seed funding round that will help it to launch and improve its automated red-teaming tools,” he said. “With the demand for AI agents, it could gain a decent share of the IT spending wallet fairly quickly.”

Since launching last August, the startup has made good progress, averaging revenue growth of 127% each quarter while onboarding customers including the professional services firm KPMG LLP, the Croatian telecommunications provider Infobip Ltd. and the generative AI search startup Glean Technologies Inc.

To accelerate growth, the startup has also open-sourced one of the key tools in its armory, called Agentic Radar, which maps the dependencies in agentic AI workflows and components, using static code analysis to expose any missing security measures within them.

Rain Capital Managing General Partner Chenxi Wang said SplxAI can become a key player in the agentic AI revolution, as it’s one of the few startups around that can effectively secure these autonomous systems.

“Organizations already recognize the benefits of investing in generative AI, but the technology is still in its infancy,” Wang said. “It’s critical we use rigorous, automated red teaming to ensure the security and reliability of these systems. SplxAI has the expertise and the technology to do this at scale.”

Image: SiliconANGLE/Microsoft Designer