Where our figures come from

Guidance of national security agencies

2. BSI (Germany) and ANSSI (France): Design Principles for LLM-based Systems with Zero Trust. August 2025. Used for: output verified by rules, human approval in critical cases. bsi.bund.de (archived copy) · cyber.gouv.fr (archived copy)
3. NSA (United States): Model Context Protocol (MCP): Security Design Considerations for AI-Driven Automation. May 2026. Used for: validation of parameters; missing approval steps named as a weakness. nsa.gov (archived copy)
4. ASD’s ACSC (Australia), CISA and NSA (United States), Cyber Centre (Canada), NCSC-NZ (New Zealand), NCSC-UK (United Kingdom): Careful adoption of agentic AI services. 1 May 2026. Used for: a second agent that validates, redundant agents, rollback and reversibility, human approval before actions with high impact. cyber.gov.au (archived copy)
5. BSI (Germany): IT-Grundschutz-Kompendium, module NET.1.1 Netzarchitektur und -design. Edition 2023. Used for: systems with different protection needs in separate network segments (A23), firewalls with allow-lists between zones (A4), physical separation for high protection needs (A31). bsi.bund.de (archived copy)

14. BSI (Germany): IT-Sicherheitsinformation 2026-262788-1032, Auswirkungen auf die Cybersicherheit von Organisationen durch die Entwicklung im Bereich Künstlicher Intelligenz. Version 1.0, 22 June 2026. Used for: security updates within hours for systems that can be reached from the internet, records and backups kept on separate systems, the check after a vulnerability that was already being exploited, the access of service providers, replacement before the end of support, and regular exercises. bsi.bund.de (archived copy)
15. BSI (Germany): AI Audit and Assurance Assessment Architecture (A5), Horizontal Trustworthiness Core Module and Cloud Infrastructure Operations Module. Community draft, version 0.9, 30 June 2026. Used for: the exit plan for the AI model, the exercise of the emergency shutdown, and the way to contest a result of the AI. A draft for comments; it is read again when the final version exists. bsi.bund.de
16. Allianz für Cyber-Sicherheit, Expertenkreis KI-Sicherheit (Germany): Informationen und Empfehlungen zum sicheren Einsatz von (generativer) KI für die Leitungsebene. Version 2.0, 24 July 2026. Used for: organisational prerequisites (AI policy, a responsible person, a list of the approved AI applications) and training material for users and for management. allianz-fuer-cybersicherheit.de (archived copy)

17. NSA (United States): Best Practices Guide for Cyber Hygiene. Version 1.0, September 2026. Used for: network segments with firewalls, phishing-resistant multi-factor authentication (MFA) for administrators, records and the detection of unusual traffic, exercised recovery plans. nsa.gov (archived copy)
18. NSA (United States): The Commercial National Security Algorithm Suite 2.0 and Quantum Computing FAQ. Version 2.1, December 2024. Used for: algorithms and key sizes, the dates for the change to quantum-safe encryption, the position on combined methods. The NSA addresses the suite to national security systems of the United States; here it is a measure chosen by ourselves. nsa.gov (archived copy)
19. NSA (United States): Zero Trust Implementation Guidelines. January 2026. Used for: mapped data flows, granular access rules, segmentation, protection of data in transit. nsa.gov (archived copy)
20. NSA and partner agencies: Improve Router Hygiene to Protect Against Russian State-Sponsored Targeting. Joint advisory, version 1.0, July 2026. Used for: the administration of routers, firewalls and VPN gateways (SNMP version 3, management from a separate network, multi-factor authentication at the central sign-in). nsa.gov (archived copy)
21. NSA and partner agencies: Deploying AI Systems Securely: Best Practices for Deploying Secure and Resilient AI Systems. Version 1.0, April 2024. Used for: origin and integrity of the files of a self-hosted model, rollback to the last good version, backups that cannot be changed. nsa.gov (archived copy)

22. BSI (Germany): IT-Grundschutz-Kompendium, modules SYS.1.1 Allgemeiner Server, SYS.1.3 Server unter Linux und Unix, SYS.1.5 Virtualisierung and SYS.1.6 Containerisierung. Edition 2023. Used for: the hardening of servers, virtual machines and containers. SYS.1.1 (archived copy) · SYS.1.3 (archived copy) · SYS.1.5 (archived copy) · SYS.1.6 (archived copy)
23. NSA (United States): Guidance for Managing UEFI Secure Boot. Version 1.0, December 2025. Used for: the check of Secure Boot on every new device. nsa.gov (archived copy)
24. NSA and CISA (United States): Harden Baseboard Management Controllers. Version 1.0, June 2023. Used for: the management controller of a server (credentials, separate network, updates). media.defense.gov (archived copy)
25. NSA and CISA (United States): Kubernetes Hardening Guidance. Version 1.2, August 2022. Used for: containers without privileges, read-only file systems and limits on system calls. media.defense.gov (archived copy)

Human oversight in organisations

13. KPMG US: AI Quarterly Pulse Survey, third quarter of 2026. Press release of 24 September 2026, and the report of the survey. Used for: 49% of the organisations surveyed have identified high-risk use cases in which autonomous decisions of AI agents are not allowed, up from 28% in the third quarter of 2025. Survey of 314 C-suite and business leaders of organisations in the United States with an annual revenue of 1 billion US dollars or more, 24 July to 25 August 2026. kpmg.com (archived copy), report (PDF) (archived copy)

The figure describes large organisations in the United States. It shows a practice, not a requirement of a standard.

AI functions of Cisco: what is available and what is announced

• Cisco, Webex blog: Cisco Live 2026: Advancing Cisco Calling with AI, Hybrid Flexibility, and Resilience. June 2026. Used for: AI Receptionist and Cisco AI Assistant for Webex Calling are generally available; both are announced for Cisco Unified Communications Manager. blog.webex.com (archived copy)
• Cisco, Webex blog: AI Receptionist for Webex Calling is Now Generally Available. June 2026. Used for: what the AI Receptionist does; further languages, German among them, are announced. blog.webex.com (archived copy)
• Cisco, Webex blog: Modernizing the on-premises Contact Center with AI; Webex AI Agent now available with CCE Release 15. August 2026. Used for: Webex AI Agent for Contact Center Enterprise 15, connected to the Cisco cloud. blog.webex.com
• Cisco, Webex Help Center: Get started with Webex AI Agent. 25 September 2026. Used for: AI agents that work on their own or along a script, by voice and by chat. help.webex.com
• Cisco, Webex Help Center: Supported languages and voices for AI agents. 30 September 2026. Used for: German and French among the languages. help.webex.com (archived copy)
• Cisco, Webex Help Center: Data Security, Privacy, and Residency in Webex Contact Center. 10 September 2026. Used for: processing of the AI functions in the EU region. help.webex.com
• Cisco, RoomOS documentation: Cisco collaboration devices with Microsoft Teams Rooms. Used for: the Cisco devices that run Microsoft Teams Rooms, their management in the Teams admin center and in Control Hub. roomos.cisco.com

Read on 13 October 2026. Announcements are plans of the vendor and can change.