National Cybersecurity
Association of Ukraine

Questions and answers.

Short answers in plain language. More detail is in the focus areas and guides.

AI security.

What is AI security?
It is about making AI safe for the people and organisations that use it, and about how AI changes attacks. The US standards body NIST names both sides: AI can lower the barrier to offensive cyber activity, and AI systems have their own weaknesses, such as prompt injection and data poisoning. We explain how to adopt AI without creating new risks for people.
What is prompt injection?
It is input that changes how an AI model behaves in ways its owners did not intend. It can be direct, written in the user’s own input, or indirect, as hidden instructions in web pages or files the model processes. The risk is reduced with least privilege for the model, human approval for risky actions, and treating external content as untrusted.
Is it safe to put company or personal data into an AI chatbot?
Not by default. Sensitive data can leak through a model’s outputs, so OWASP advises telling users not to enter sensitive information and publishing retention and deletion policies. Set a written rule on what may be entered, and check the provider’s terms and the data-protection rules that apply to you before using personal data.
What is “excessive agency”, and why do AI agents worry security teams?
Harm becomes possible when a model’s unexpected or manipulated output can trigger actions, because it was given too many tools, permissions or autonomy. The 2026 joint guidance from government cybersecurity agencies advises limiting access and starting with low-risk uses.
Which frameworks should a small organisation follow?
Start with the OWASP Top 10 for LLM applications as a checklist, the NIST AI RMF (Govern, Map, Measure, Manage) for structure, and the 13 principles of the UK Code of Practice for the Cyber Security of AI as a practical baseline. None of them is a certificate, and the UK code is voluntary.
Can an AI’s wrong answer become a security problem?
Yes. NIST lists “confabulation” among its 12 generative-AI risks, and OWASP lists misinformation, meaning false but credible output with legal and reputational consequences. For high-stakes use, require a human check and ground answers in trusted sources.
How do I judge whether an AI vendor is secure enough?
Ask how the vendor handles and retains data, where its models and components come from, how it patches, and how it tests and monitors the system. OWASP’s supply-chain item advises vetting suppliers, keeping an inventory of components and checking provenance, and the UK code has a supply-chain principle. A list of questions like this is our suggestion, not a standard.
What is data poisoning?
It is tampering with the data used to pre-train, fine-tune or feed a model so that it learns hidden flaws or backdoors. Defences include tracking where data comes from, vetting data suppliers and checking for anomalies. CISA and its partners also publish AI data-security guidance.

Deepfakes.

What is a deepfake?
It is synthetic audio, video or imagery made with AI to look real, such as a face swap or a cloned voice. NIST describes realistic synthetic audiovisual content and photorealistic images. Cheaper edits of real media are a different problem with similar defences.
Can I tell a deepfake by eye?
Not reliably. A Ukrainian state centre reported in November 2025 that some AI-generated videos cannot be debunked without additional verification tools, and the FBI lists visual glitches as only one cue among several. Check the source and the context first.
Can an AI detector tell me whether a video is fake?
No tool should give a verdict. In a 2025 study of in-the-wild content from 2024, open-source detectors lost roughly half of their AUC compared with earlier benchmarks, and reviews of detection tools point to false positives, language gaps and variable accuracy. Treat a result as one input, with its confidence and limits stated.
Someone made a deepfake of me. What do I do first?
Don’t spread it and don’t pay anyone. Preserve the original, the page addresses, the timestamps and a hash value of the file, then report to the platform and to the authorities. The guide “The first 24 hours after a deepfake” sets this out step by step.
How do I protect my family and finance team from voice-clone calls?
Agree a secret phrase, hang up and call back on a number you look up yourself, and never pay under pressure by wire transfer or gift card. For companies it makes sense to add callback verification and two-person approval for payments. That last part is our suggestion, not a standard.

Cyber warfare and hybrid threats.

What is cyber warfare, and how does it differ from ordinary cybercrime?
In public sources the difference is mainly who acts and why. ENISA’s threat picture separates state-aligned espionage, ransomware and other crime, and hacktivism. Naming specific actors is a job for the authorities, not for NAK explainers.
What are hybrid threats?
They are a mix of cyber operations, information manipulation and pressure used together. The EU’s diplomatic service (EEAS) describes foreign information manipulation and interference (FIMI) as manipulative, intentional and coordinated behaviour that is mostly not illegal, including disinformation. We focus on defending against the behaviour itself.
Are small organisations and civilians really targeted?
ENISA’s data for the EU shows phishing as the dominant way in (60%), and hacktivist activity making up almost 80% of recorded incidents, mostly low-impact DDoS. That is a European picture, not a Ukrainian one: we do not quote Ukrainian figures until we have read a primary report.
What should a small organisation do first?
Back up, including an offline copy, and test your restores; turn on multi-factor authentication; patch; brief staff on scams; and write down who to call. These steps match the UK NCSC’s advice for small businesses and CISA’s ransomware guide.
What is a DDoS attack and should I worry?
It is flooding a service with traffic so that it becomes unavailable. ENISA says hacktivist DDoS dominates incident counts but usually has very low impact. If you run a public website, ask your hosting provider about DDoS protection and plan in advance how you will communicate during an outage.
Where do I report a cyber incident in Ukraine?
Organisations can write to CERT-UA at incidents@cert.gov.ua, attaching logs and files after isolating the network; CERT-UA is not a punitive body. Individuals can use the Cyber Police’s report form or helpline. Check the contacts on the official sites before sending anything.
How can I tell an information operation from an ordinary rumour?
Look at behaviour: coordination, intent to manipulate, repetition across accounts. The Ministry of Digital Transformation gives an example of an official-looking fake site that started panic about “summonses through Diia” and had no link to any authority. You cannot prove coordination from a single post, so do not name accounts.

About NAK.

Is NAK a government body?
No. NAK is an independent association registered in Kyiv on 15 January 2020. It is not part of the National Security and Defense Council, its National Cybersecurity Coordination Center or any other state body.
What does NAK do?
It brings together the state, business, the expert community, researchers and technology companies to build a safe, resilient and open digital environment: developing public-private collaboration, knowledge sharing and joint initiatives, and publishing practical, sourced material.
How is NAK different from the “Cybersecurity Association of Ukraine”?
They are different organisations with different registry codes: NAK is 43452536, the Cybersecurity Association of Ukraine is 43289798.
Does NAK check files for authenticity or remove content?
No. NAK publishes methods and practical guides, but it does not examine files or remove material from platforms. What to do is set out in the guide “The first 24 hours after a deepfake”.

What the answers rely on