iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
AI is not simply a bane to cybersecurity. It can help defenders, help attackers work at scale, and expose organizations to new risks when they deploy AI systems. Official guidance and threat reporting document these effects, but do not establish that AI makes cybersecurity worse overall.
How AI changes cybersecurity
AI is a dual-use technology: the same broad capability to analyze information or generate convincing content can support defensive work or malicious activity. The U.S. National Institute of Standards and Technology (NIST) says AI may augment cybersecurity defenses while also enhancing the capabilities of people targeting organizations and individuals.
It helps to separate three questions: how attackers use AI, how AI systems themselves can be attacked, and how organizations might use AI to defend other systems. Evidence of one does not settle the others—or prove AI’s overall net effect.
How attackers can use AI
Phishing and social engineering
The European Union Agency for Cybersecurity (ENISA) reported in its 2025 threat-landscape release that large language models were being used to enhance phishing and automate social-engineering activity. These tools can help create or adapt messages, but that does not mean every phishing attempt is AI-generated.
#1 Best Overall
ENISA also reported that, by early 2025, AI-supported phishing campaigns represented more than 80% of observed social-engineering activity worldwide. Treat that as a figure reported by ENISA, not as an independently verified global census: the release does not provide a primary measurement study for the claim in the material available here.
Synthetic media and broader distribution
In its 2026 release, ENISA described AI-generated audio, video, and text being used in information manipulation. It also said AI can enable mass distribution and translation. These capabilities can make impersonation or misleading content easier to adapt for different audiences, but they do not establish that any particular message, recording, or video is synthetic.
For people and organizations, the practical risk is that convincing presentation is not reliable proof of identity. Handle sensitive requests—especially requests to transfer money, share credentials, or change account details—through a separate, trusted verification process.
How AI systems can become targets
Putting AI into a product or business process adds components and data flows that need protection. NIST identifies confidentiality, integrity, and availability risks involving AI systems, their training and output data, and the software and hardware they rely on. These systems also inherit familiar security weaknesses; AI does not replace the need to secure ordinary infrastructure and applications.
Rank #3
NIST notes that existing frameworks and guidance do not comprehensively address some AI-related attacks, including evasion, model extraction, membership inference, and availability attacks. Exposure depends on the system, its data, its deployment, and its use; these concerns do not mean every AI system is equally vulnerable.
What adversarial machine-learning attacks mean
NIST’s Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations (AI 100-2e2025), published on March 24, 2025, groups attacks by how they affect predictive and generative AI systems. It is voluntary guidance, not a security certification or a guarantee that a system is safe.
| Attack class | Plain-language meaning | Where NIST includes it |
|---|---|---|
| Evasion | An attacker manipulates inputs to change a system’s output. | Predictive and generative AI |
| Poisoning | An attacker corrupts training or learning data. | Predictive and generative AI |
| Privacy attacks | An attacker seeks to infer or expose information. | Predictive and generative AI |
| Misuse | A generative system is used for harmful purposes. | Generative AI |
NIST discusses mitigations as well as their limitations. A claim that a product or control “secures AI” is therefore too broad to evaluate on its own. Ask which threat class it addresses, for which system and use, and what evidence supports its effectiveness.
How AI may support cybersecurity defenses
NIST recognizes that AI may augment defensive capabilities. That is a potential use, not a guarantee that an AI detection tool will improve security in every environment or replace expert review. Defenders also need to account for AI-enabled offensive techniques and protect the AI components they rely on.
Best Value
When comparing a defensive measure, distinguish between using AI to defend other systems and protecting an AI system itself. Also check which components and data are in scope, which attack classes the measure addresses, whether its support comes from voluntary guidance or measured operational evidence, and what human review or other controls remain necessary.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the reported numbers do—and do not—show
ENISA’s 2025 Threat Landscape analyzed 4,875 incidents observed from July 1, 2024, through June 30, 2025. The publication page records a version 1.3 revision dated September 22, 2026, correcting figures and links. The count describes ENISA’s defined incident dataset; it is not a count of incidents caused by AI.
ENISA’s 2026 release covers events observed from January 1 through December 31, 2025, drawing on open-source reporting and anonymized information shared by EU Member States and through ENISA’s Cyber Partnership Programme. ENISA also reported more than 48,000 newly assigned CVE identifiers in 2025, a 22% increase from the prior year. That is a general vulnerability count—not a count of AI-specific vulnerabilities or attacks.
Taken together, these figures and reports show that AI is being used in some threat activity and that AI deployments create security exposure. They do not quantify AI’s total causal contribution to cybercrime or establish a universal net harm-benefit result.
Practical steps for organizations
- Inventory AI use. Identify AI tools and AI-enabled services, what data they handle, and the components and dependencies they rely on. Include third-party services as well as systems built or operated in-house.
- Protect the surrounding systems. Apply established cybersecurity and secure-development practices to the software and infrastructure around AI, then assess risks specific to the system’s data, model, and use case.
- Name the threat before choosing a control. Determine whether the concern is evasion, poisoning, privacy, misuse, or another defined risk. Check whether a proposed measure addresses that threat in the organization’s actual deployment and what its evidence establishes.
- Verify consequential requests independently. Treat AI-assisted messages, voice, and video as possible impersonation channels. Use a separate trusted process to confirm requests involving money, credentials, or sensitive changes rather than relying on how polished or authentic a message appears.
- Keep attribution precise. Do not treat broad incident or vulnerability totals as proof of AI causation unless the source explicitly makes that attribution.
So, is AI a bane to cybersecurity?
Calling AI a bane overstates what the available evidence establishes. AI can make some malicious workflows more scalable or persuasive, and AI deployments introduce additional attack surfaces. It may also augment defensive work. The sound conclusion is not that AI is inherently harmful or inherently protective, but that its security impact depends on how it is used, what is exposed, and whether defenses address the relevant threats.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
