HackerOne is expanding its AI security offerings with the introduction of Agentic Prompt Injection Testing, a new approach designed to determine whether AI systems can be exploited in production environments. The move comes as organizations face a sharp rise in prompt injection vulnerabilities, a growing threat tied to the widespread adoption of large language models.
According to HackerOne, validated prompt injection issues on its platform have increased by 540 percent over the past year. The surge underscores how quickly attackers are adapting to AI-powered systems, particularly those connected to enterprise data and automated tooling.
Unlike traditional application vulnerabilities, prompt injection attacks target how AI models interpret and act on input. When these systems are integrated with internal databases, APIs, and workflows, the consequences extend beyond incorrect outputs. A successful attack can lead to unauthorized data access or trigger actions that were never intended by system designers.
This shift is exposing limitations in existing AI security strategies. Many defenses focus on screening inputs at the model level, aiming to block malicious prompts before they are processed. However, these measures often fail to capture how AI systems behave under sustained or indirect attack scenarios.
Agentic Prompt Injection Testing is built to address that gap. The capability simulates adversarial behavior against live AI deployments, using multi-step interactions to test whether vulnerabilities can be exploited in practice. It also evaluates how attacks propagate through connected components such as retrieval systems and external tools.
By focusing on real-world outcomes rather than theoretical weaknesses, the testing provides security teams with evidence of actual exposure. The findings are designed to be reproducible and prioritized by severity, helping organizations move from detection to remediation more efficiently.
HackerOne developed the new capability through its work with companies including Anthropic, IBM, Snap, Adobe, and eBay, where securing AI systems in production has become an immediate priority.
“Prompt injection has quickly become a severe risk to deployed AI systems because it can transform a trusted application into an attack surface,” said Nidhi Aggarwal, Chief Product Officer at HackerOne.
“Security teams can’t rely on static controls or runtime filters alone. They need validated proof of whether an AI system can be exploited once it’s connected to real data and tools. Agentic Prompt Injection Testing delivers that evidence, enabling organizations to identify confirmed exposure and reduce risk before it impacts the business.”
“AI is advancing at an unprecedented pace, and security must keep up,” said Omar Santos, Distinguished Engineer, AI Security at Cisco and Project Governing Board Co-Chair of the Coalition for Secure AI (CoSAI). “The future leading organizations will be defined not just by AI innovation, but by how well they secure it. Rigorous validation under adversarial conditions is becoming fundamental to building trustworthy AI systems and raising the bar for the industry.”
The release positions adversarial testing as a necessary layer in AI security, particularly as systems move from experimentation into critical business functions. Rather than relying solely on preventative controls, organizations are beginning to prioritize validation methods that reflect how attacks actually unfold.
As AI adoption accelerates, the ability to test systems under realistic conditions is emerging as a key factor in reducing operational and security risk.
