Recent security incidents at well-known AI vendors show just how much generative AI is transforming the IT security landscape. Juraj Janosik, Vice President of AI at ESET, assesses the situation and calls for binding protective mechanisms from AI providers.
The European IT security vendor ESET, through its Vice President of AI, Juraj Janosik, points to recent security incidents at well-known AI vendors as evidence that generative AI systems are increasingly becoming attack vectors and risk factors in their own right. The source does not provide specific details on individual incidents, but emphasizes that the accumulation of such events is increasing the pressure on providers to act.
For CISOs, this shifts the risk assessment: AI products integrated into enterprise processes bring not only classic vulnerabilities but also new attack surfaces through the models themselves — for example, through prompt manipulation, misbehavior in edge cases, or insufficiently tested training data. According to Janosik, anyone who tests or approves such systems also bears responsibility for the consequences of misbehavior in production operations. This affects not only manufacturers but also companies that evaluate AI systems and integrate them into their own environments.
The central demand is that AI companies must build appropriate protective mechanisms (“guardrails”) into their products from the outset to prevent them from becoming a source of danger themselves. For security officers within companies, this means paying increased attention to documented security tests, liability clauses in contracts, and clear responsibilities in the event of malfunctions when selecting and approving AI tools, before systems are transferred into production environments.
ESET expert Juraj Janosik, in light of recent security incidents at AI vendors, is calling for binding protective mechanisms and clear liability rules for testers and providers of AI systems.
Source: itwelt.at · Published August 3, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrasing and classification by Lumi News Pipeline v1.8.3.