A10 Networks has announced the acquisition of TrojAI Inc., a company focused on AI model security and protection against adversarial attacks. The move is part of A10’s broader strategy to expand its presence in the growing AI security market.
TrojAI specializes in testing and hardening AI and machine learning models against a range of threats, including adversarial inputs, model poisoning, and extraction attacks. Its technology helps organizations identify vulnerabilities in AI models before and after deployment. By acquiring the company, A10 aims to combine its existing application and infrastructure security strengths with deeper AI-specific protections.
The acquisition comes as more enterprises deploy AI models into production environments, where they become exposed to new types of attacks that traditional security tools were not designed to handle. While many organizations have focused on securing the infrastructure around AI, fewer have invested in securing the models themselves.
A10 Networks, traditionally known for application delivery controllers and DDoS protection, has been gradually expanding into areas that support AI workloads. The addition of TrojAI’s capabilities gives A10 a more complete offering for customers concerned about the security of their AI systems.
This deal reflects a wider trend of established cybersecurity vendors acquiring specialized AI security startups to quickly build out their AI defense portfolios. Rather than developing these capabilities internally, companies like A10 are choosing to acquire targeted expertise and technology.
Conditions Driving the Change
Several factors are driving increased investment in AI model security:
Enterprises are deploying more AI models into production, increasing their exposure to targeted attacks.
Adversarial attacks on AI models have moved from theoretical research into real-world threats that can manipulate model outputs or steal intellectual property.
Traditional cybersecurity tools were built to protect networks and applications, not the underlying AI models themselves.
Regulatory pressure around AI is growing, with organizations needing to demonstrate that their models are secure and reliable.
Many companies lack internal expertise in AI-specific threats, creating demand for specialized security solutions.
High-profile incidents involving manipulated AI systems have raised awareness of model-level risks among security leaders.
The rapid growth of generative AI and autonomous agents has expanded the attack surface beyond traditional machine learning models.
Security teams are being asked to secure AI systems without clear frameworks or tools designed for this purpose.
Vendors are recognizing that AI security is becoming a distinct category rather than a feature of existing security platforms.
Acquisitions are accelerating as larger cybersecurity companies seek to close capability gaps quickly instead of building solutions from scratch.
These conditions have made AI model security a strategic priority for both vendors and enterprise security teams.
What AI Security Looked Like Before
Before specialized AI model security companies gained traction, most organizations approached AI security through a traditional cybersecurity lens. Security teams focused on protecting the infrastructure, networks, and data pipelines that supported AI systems. This included securing cloud environments, managing access controls, and monitoring for general cyber threats.
However, the models themselves were often treated as black boxes. Once a model was trained and deployed, there was limited visibility into how it could be attacked or manipulated. Few organizations had processes to test models against adversarial inputs or to detect if a model had been compromised after deployment.
Security tools at the time were largely reactive. They could detect unusual network traffic or application behavior, but they were not designed to understand or defend against attacks that targeted the logic inside an AI model. As a result, many enterprises had significant blind spots when it came to model integrity and robustness.
Governance around AI security was also inconsistent. While some highly regulated industries conducted basic model risk assessments, most organizations lacked structured approaches to securing AI systems throughout their lifecycle. Security was often an afterthought once models moved into production.
Overall, AI security before this wave of specialized tools was fragmented and infrastructure-focused. It did little to address the unique vulnerabilities that exist inside AI models themselves.
What AI Security Looks Like Now
AI security has evolved into a more specialized discipline that includes direct protection of models, not just the systems around them. Organizations are increasingly adopting tools that can test models for vulnerabilities, monitor them for signs of compromise, and apply defenses against adversarial attacks.
Security teams now recognize that AI models can be attacked in ways that traditional cybersecurity tools cannot detect. This has led to greater investment in capabilities focused on model robustness, adversarial testing, and runtime monitoring of AI behavior.
The market has also shifted toward integrated approaches. Rather than treating AI security as a completely separate silo, many organizations are looking for solutions that combine infrastructure security with model-level protections. Acquisitions like A10’s purchase of TrojAI reflect this trend, as vendors aim to offer more comprehensive coverage.
Another change is the growing emphasis on continuous security rather than one-time assessments. Organizations are beginning to view AI model security as an ongoing requirement, similar to how they manage vulnerabilities in software and infrastructure.
While adoption is still uneven across industries, the direction is clear: AI security is moving from a niche concern to a core part of enterprise security programs, especially as AI systems take on more critical and autonomous roles.
Our Take
AI Security Take
The acquisition of TrojAI by A10 Networks shows that established cybersecurity vendors are now treating AI model security as a necessary part of their portfolio. As more enterprises run AI in production, simply securing the surrounding infrastructure is no longer sufficient.
For security leaders, this means expanding their scope to include model-level threats. This includes testing models for adversarial vulnerabilities, monitoring for signs of tampering or extraction, and implementing controls that can detect when an AI system is behaving unexpectedly.
However, acquisitions alone will not solve the broader challenge. Many organizations still lack clear ownership of AI security within their teams. Security, data science, and compliance groups often operate in silos, making it difficult to implement consistent protections across the AI lifecycle.
Teams evaluating solutions in this space should focus on how well any new capabilities integrate with their existing security stack and whether they provide actionable visibility rather than just additional alerts. The real value will come from tools that help organizations understand and reduce risk without creating excessive operational overhead.
As AI adoption continues to grow, the gap between organizations that actively secure their models and those that do not will likely become more visible — both in terms of resilience and regulatory readiness.
