Executive Summary
- 17% of companies report confirmed or suspected incidents in which AI played a role.
- Only 11% of surveyed companies have specific procedures for AI‑related security incidents.
- Common attack types: convincing phishing emails (90%), automated attack scripts (40%), deepfakes (11%).
- 49% already use AI, 9% plan to deploy it within twelve months; only 31% of AI users apply AI to improve IT security.
- 69% expect greater security from independent audits; 51% view certifications as an important selection criterion.
What happened?
Facts: A joint study by the TÜV Association and the Federal Office for Information Security (BSI), based on more than 500 surveyed companies with 20 or more employees, documents the current state of AI and cybersecurity. In total, 17% of companies report in the past twelve months confirmed incidents (4%), suspected incidents (11%) or both (2%) in which AI played or may have played a role. Only 11% of companies have specific procedures for handling AI‑related security incidents. The most commonly reported attack types are convincing phishing emails (90% of affected companies), automated attack scripts (40%) and deepfakes (11%).
Further facts from the survey: 49% of companies use AI, 9% plan to do so within the next twelve months, and 42% currently do not use AI and do not plan to. Among companies using AI, only 31% use AI to improve IT security. Common protective measures for AI applications are quality assurance and human review (67%), restricted access rights (64%), protection of sensitive data (62%), testing before deployment (52%) and logging/monitoring (40%). 8% reported having implemented none of the listed technical measures.
Why does it matter?
Facts: The study identifies both external threats from AI‑assisted attacks and risks arising from one’s own AI use. 69% of companies that use or plan to use AI expect greater security from independent audits; 51% consider certifications by independent testing organizations an important selection criterion. The TÜV organizations are working together with the TÜV AI.Lab on a framework program for certifying AI systems, being piloted with partners and expected to be available by the end of the year. Supporting materials such as the handout from the Alliance for Cyber Security (ACS) for mid‑sized companies are available.
Assessment: For decision‑makers this means the threat landscape is not just theoretical: the survey documents concrete incidents and widespread attack patterns. At the same time, the study reveals gaps in processes and technical measures, both in defense and in securing internal AI use. The interest in independent audits and certifications indicates a demand for reliable evidence of AI application trustworthiness.
What does this mean for mid-sized companies?
Facts: Many mid‑sized companies cite data protection and security concerns as reasons against AI (53%), followed by lack of perceived benefit (58%), high integration effort (44%), legal uncertainties (43%), lack of skills (39%) and expected costs (36%). Among companies using AI, measures such as employee training (56%), risk assessments (45%), formal approval processes (45%) and supplier reviews (42%) are already common, but 29% have implemented none of these measures.
Assessment: The figures show that mid‑sized companies are simultaneously targets and potential sources of risk: external attackers use AI for automation and deception, while unsecured internal AI use can expose data or cause wrong decisions. For decision‑makers this means specifically: lacking AI‑specific processes increases the risk of overlooking incidents or responding incorrectly; insufficient technical protections reduce the effectiveness of defensive strategies.
Related: AI Readiness Check
Recommendation from Future AI Solutions
Based on the study results we recommend pragmatic, immediately implementable steps — presented separate from the facts:
- Augment processes: Add AI‑specific scenarios to existing IT security and incident‑response processes (detection, reporting channels, escalation). The study shows only 11% have such procedures.
- Implement technical protections: Prioritize quality assurance and human review of critical AI outputs, restrict access rights, anonymize sensitive data and run tests before production deployment — measures that parts of the surveyed companies already apply.
- Expand monitoring and logging: Only 40% log or monitor AI usage; extended logging facilitates detection and forensic analysis.
- Assess suppliers and products: Conduct supplier evaluations and request evidence of security; 51% of AI users consider certifications important.
- Training and approval processes: Train staff and establish formal approvals for AI applications (already implemented by some companies).
- Consider independent audits: Use available offers, the ACS handouts and the forthcoming audit/certification frameworks as decision aids; many companies expect increased security from these.
These recommendations are intended as orientation. Concrete measures should be tailored to company size, use case and risk exposure.