The point: AI security depends on governance and human accountability, not model size; increased computing power without organizational control only amplifies existing weaknesses.
AI systems are becoming more powerful, but their security does not depend on computing power, but rather on whether organizations can understand, contextualize and take responsibility for results. Matt Wood, Head of Cyber Security Operations at ABBYY, warns: governance and human accountability are decisive, otherwise more performance only amplifies existing weaknesses.
The central security challenge with AI does not lie in the technical performance of models, but in the organizational ability to understand and take responsibility for AI decisions. In security-critical areas such as cybersecurity, AI outputs have immediate operational consequences. While AI can detect anomalies faster and make risks visible earlier, it does not relieve organizations of the responsibility to clearly define objectives, critically review results, and bear the consequences of decisions.
A common misconception is that more computing power, larger models and faster hardware automatically solve security problems. In reality, they can even amplify existing weaknesses: faulty objectives, inadequate data sources or loose controls are not fixed by higher speed, but rather enable the same mistakes to be made at greater scale and speed. AI results often appear convincing, well-structured and precisely formulated, which leads to them being confused with reliable truth. Teams risk optimizing the wrong thing or overlooking risk developments when they incorporate answers into operational processes without adequately validating them.
The real risk does not arise from model size, but from a discrepancy between technical performance and organizational governance: if computing power scales faster than understanding and accountability, security gaps emerge. AI is indeed strong at pattern recognition, but secure decisions require context, clear boundaries and human judgment. A system can produce professionally plausible answers, yet still create risks in practical deployment – for example by merely shifting a problem rather than solving it.
From a security perspective, pattern recognition alone is not a sufficient security model. It is a useful capability within a broader system of governance, clear accountability and human oversight. The human must decide which insights are relevant, which risks are acceptable and when an AI answer must be questioned. In cybersecurity, it ultimately comes down to consequences – and taking responsibility for them is the organization’s task, not the algorithm’s.
Source: www.it-daily.net · Published 13 July 2026
Lumi AI News — AI-assisted curation pursuant to Article 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.3.