AI security training surges as prompt injection tops list

Hack The Box has published a Cybersecurity Workforce Intelligence Report on AI-related security training trends, drawing on anonymised activity from more than 702,000 cybersecurity professionals worldwide.

The findings point to a sharp rise in training around AI attack methods as security teams adapt to risks linked to prompt-based systems, machine learning models and autonomous software agents.

Prompt injection was the most common AI-focused area in solved challenges, accounting for 29% of the dataset. Machine learning model exploitation made up 24%, including model backdooring, weight extraction and supply-chain compromise scenarios, while agentic AI hijacking represented 12%.

These patterns suggest practitioners are moving beyond conventional security disciplines and preparing for attack routes tied directly to the design and deployment of AI systems. They also indicate that employers must rethink how they train teams as AI becomes more embedded in business operations and security work.

The data covers professionals across 251 countries and territories. It also shows cybersecurity upskilling becoming more geographically dispersed, with India identified as a major talent centre alongside the United States, the United Kingdom, France and Brazil. Together, those markets accounted for nearly 36% of the global upskilling activity captured in the study.

Blurring roles

The data also shows growing overlap between offensive and defensive training. According to the report, 77.53% of defensive professionals also pursued offensive security training, suggesting specialists are increasingly working across both areas rather than staying within narrower role definitions.

That trend reflects a broader move towards what the industry often calls a purple-team model, in which attack and defence skills are developed together. In practice, organisations are placing more value on staff who can test systems, identify weaknesses and understand how those weaknesses might be exploited.

Structured learning programmes are also playing a larger role in this shift. AI-focused training completion rates reached 64%, underlining the impact of organisation-led programmes in developing more advanced security skills.

That matters for employers facing a shortage of experienced cyber staff and a rapidly changing threat landscape. More firms are trying to build talent internally rather than rely solely on recruitment, especially in areas where AI expertise remains scarce.

Hack The Box presents the findings as a workforce planning issue as much as a technical one. Security leaders, it argues, need to prioritise AI security skills, combine offensive and defensive development, widen international talent pipelines and maintain continuous hands-on training.

Haris Pylarinos, founder and chief executive officer of Hack The Box, said the divide between teams that can use AI effectively and those that cannot is becoming a direct source of risk.

"AI is creating a divide between teams that can operationalize it and those that can't, and that divide directly translates into risk," Pylarinos said. "For CISOs, the challenge is ensuring their teams can operate effectively with AI, and without it when needed."

Training demand

The report identifies AI penetration testing as one of the main global training priorities. That suggests organisations are no longer treating AI security as a niche topic, but are starting to build it into routine security testing and workforce development.

The emphasis on hands-on training also indicates that employers want practical evidence of skill development, not just theoretical knowledge. Simulated exercises and challenge-based learning have become common ways to test whether staff can recognise and respond to new forms of attack.

For chief information security officers, the findings present a management challenge that goes beyond buying software tools. Teams need to assess AI systems, validate outputs, test model behaviour and defend environments that may include both traditional infrastructure and AI-driven services.

The changing mix of skills could also reshape cybersecurity career paths as practitioners broaden their expertise. Rather than separating red-team, blue-team and specialist AI roles too rigidly, employers may need more flexible team structures built around adaptable staff with cross-domain knowledge.

With prompt injection, model exploitation and agentic AI hijacking leading the AI-related challenges solved by users in Hack The Box's data, the study suggests cybersecurity training is already shifting to reflect how AI systems are being targeted in live environments.