Dieser Artikel ist derzeit auf Englisch verfügbar.
AI Governance for Cybersecurity Organisations: Using AI Responsibly, Defending Against AI Threats
Cybersecurity organisations face a dual governance challenge: governing their own use of AI in threat detection, incident response, and security products; and advising clients on AI-specific cybersecurity threats. Both dimensions create specific governance obligations.
Key Takeaways
AI used in cybersecurity products — threat detection, anomaly detection, automated incident response — is generally not classified as high-risk under EU AI Act Annex III, but transparency and accuracy obligations apply.
AI cybersecurity tools that process personal data (user behaviour analytics, SIEM with user context) are subject to GDPR and sector data protection requirements alongside any AI-specific obligations.
Adversarial AI is a real threat: threat actors are using AI to generate more convincing phishing, automate vulnerability discovery, and bypass AI-powered detection systems. Security governance must address AI-vs-AI threat scenarios.
For managed security service providers (MSSPs), AI in client environments creates additional governance complexity — MSSP AI tools that process client data have supply chain compliance implications under EU AI Act and GDPR.
NIST AI RMF (US) and ENISA AI cybersecurity guidelines (EU) provide the primary governance frameworks for cybersecurity AI — organisations should map their practices against both.
"Nur zu Informationszwecken. Dieser Artikel stellt keine rechtliche, regulatorische, finanzielle oder professionelle Beratung dar. Konsultieren Sie einen qualifizierten Spezialisten für spezifische Beratung."
The cybersecurity sector's AI governance double challenge
Cybersecurity organisations face AI governance challenges on two fronts simultaneously: governing their own use of AI in security products and operations; and preparing to advise and assist clients on AI-specific security risks. Both require structured governance approaches, and the two dimensions interact in complex ways — cybersecurity companies that use AI tools with governance gaps are poorly positioned to advise clients on AI governance.
AI in security products: what the regulations require
Cybersecurity AI products — endpoint detection and response (EDR) tools with ML-based threat detection, SIEM platforms with anomaly detection, network traffic analysis AI, automated incident response systems — are generally not classified as high-risk under EU AI Act Annex III. They are not making decisions about individuals in consequential sectors like employment or credit. The EU AI Act's transparency and limited risk provisions apply, but not the full Annex III conformity assessment regime.
However, cybersecurity AI that processes personal data has a different compliance profile. User and entity behaviour analytics (UEBA) systems that monitor individual employee activity, AI-powered insider threat detection, and identity analytics tools all process personal data — creating GDPR obligations that layer on any AI-specific requirements. Purpose limitation, data minimisation, and transparency requirements of GDPR apply to employee monitoring AI just as to customer-facing AI. In Germany and other EU jurisdictions with works council rights, deploying monitoring AI in enterprise environments also requires employee representative consultation.
Adversarial AI: the threat the governance frameworks don't fully address
The most distinctive challenge for cybersecurity AI governance is adversarial AI — the use of AI by threat actors to attack AI systems, evade AI detection, and generate more effective cyberattacks. Three adversarial AI threat patterns are now well-documented: data poisoning (corrupting training data to degrade model performance or introduce backdoors); evasion attacks (crafting inputs that cause AI classifiers to misclassify — malware designed to evade ML-based detection); and model extraction (reconstructing a target AI model through query patterns to understand its decision logic and craft bypasses).
Existing AI governance frameworks — GDPR, EU AI Act, sector-specific regulations — were not designed with adversarial AI specifically in mind. NIST's AI Risk Management Framework (NIST AI RMF) addresses adversarial ML as a risk category, and ENISA (EU Agency for Cybersecurity) has published AI cybersecurity guidelines that address adversarial attack scenarios. Cybersecurity AI governance should explicitly incorporate adversarial robustness testing, model protection controls, and anomaly detection for AI system behaviour as standard governance practices.
Managed security service provider implications
MSSPs that use AI tools in client environments face a supply chain governance dimension. AI tools processing client data are within scope of EU AI Act deployer obligations — the MSSP is the operator responsible for human oversight, monitoring, and impact assessment even when using third-party AI security tools. Client contracts should clearly delineate responsibility for AI governance between MSSP and client, and MSSP governance programs should cover AI tools deployed in client environments as well as internal operations.