# Warning: where the AI cyber threat is heading

By Ivo Donker · 7 November 2026 — written with AI assistance (Claude Opus 5.5); facts checked against the sources listed below

The attacker of 2026 is not smarter than the one from 2022, but is much faster. AI shortens the time between a published vulnerability and an attack, makes convincing fraud cheap, and adds a whole new attack surface with AI agents. This outlook sets out what the major threat reports are measuring now, what that means for 2027 to 2030, and why 'we have a back-up' is no longer enough.

## What the figures now show

For the first time in nineteen years, a data breach according to the Verizon Data Breach Investigations Report 2026 more often begins with the exploitation of a vulnerability (31%) than with stolen login credentials. Mandiant reports in M-Trends 2026 for the sixth year in a row that exploits are the main access route (32%). The figures differ because the studies measure different organisations, but the direction is the same: anyone with an outdated service exposed to the internet is once again the easiest target.

Speed is the real news. Mandiant measures that attackers sell on or hand over access once obtained within a median of **22 seconds** to a second group; in 2022 that was still more than eight hours. CrowdStrike sees in its Global Threat Report 2026 that criminals need an average of **29 minutes** to move from the first infected computer to the rest of the network; the fastest case took 27 seconds. These are different measurement points, so do not add them together. Together they do say: a person looking at an alarm in the morning is too late.

The Dutch picture is also grim. The [Cybersecurity picture Netherlands 2026](https://www.ncsc.nl/) (NCSC and NCTV, 6 November 2026) concludes that malicious actors demonstrably use AI in all phases of an attack and that the digital resilience of Dutch organisations is lagging behind the threat. The SME sector and small organisations have the least room to catch up.

## AI as accelerator and as target

CrowdStrike reports 89 % more operations in which attackers use AI and saw malicious prompts in more than ninety organisations. That is a measurement from a single security vendor, so read it as indicative rather than a worldwide tally. More important is the pattern that all sources share:

- **Exploration at scale.** An AI agent can fingerprint thousands of systems (which software, which version) and keep that map until a usable vulnerability appears. How it works, read in [how AI attacks your systems](https://cyber.llmnet.nl/en/hoe-ai-aanvalt).

- **Fraud that sounds like your boss.** A voice clone needs only seconds of audio. Deep‑fake video calls have already been used to swindle tens of millions.

- **AI itself as an attack surface.** Agents that read e‑mail, call tools and execute code can be misled with text in a web page or document (prompt injection). The OWASP Top 10 for agentic applications (December 2025) lists, among others, goal hijacking, misuse of tools, overly broad privileges, compromised agent memories and “rogue agents”.

- **The supply chain.** According to Verizon, a third party plays a role in 48 % of data breaches. An agent that “just” installs a package, plug‑in or MCP server effectively brings a new supplier with execution rights into the environment.

IBM counted in the Cost of a Data Breach Report 2025 that 13 % of the surveyed organisations reported a breach of an AI model or AI application; of that group, 97 % had inadequate access controls on that AI. The average cost of a data breach worldwide was $4.44 million.

## Projected to 2027–2030

Cybersecurity predictions date quickly, so below are no prophecies but developments already underway, supported by policy or measurement data.

| Period | What lies ahead | What it requires of you |
| --- | --- | --- |

| 2026–2027 | AI accelerates reconnaissance, phishing and the exploitation of vulnerabilities. The Cyber Resilience Act has required manufacturers since 11 September 2026 to report actively exploited vulnerabilities within 24 hours; the Cybersecurity Act (NIS2) has applied in the Netherlands since 15 August 2026. | Know what you have running, phishing-resistant login (passkeys), fixed patching deadlines, a reporting process. |

| 2027–2028 | Agents and MCP connections are becoming commonplace, as are attacks via that layer. The heavier CRA obligations apply from 11 December 2027; the AI Act rules for high-risk systems move to December 2027 and August 2028. | A separate identity and limited rights per agent, a list of permitted tools, human approval for money, data and irreversible actions. |

| 2029–2030 | ENISA places compromised software dependencies at number one of 21 emerging threats in its outlook to 2030. NIST plans the move to post-quantum cryptography for critical systems around 2030–2031. | Signed and traceable software, recovery that attackers cannot erase, an inventory of your cryptography. |

One scenario to make the scale tangible, explicitly as a worked example and not a prediction: if the number of AI-driven attack operations continues to grow by the 89% measured by CrowdStrike each year, an index of 100 in 2026 stands at 189 in 2027 and 357 in 2028. The lesson is not the number, but the direction: defence must automate now, not only once the attack is already fully automated.

## The new target: your recovery

Ransomware groups no longer just encrypt. Mandiant describes attacks that first disable backups and the recovery environment ("recovery denial") and only then strike. Microsoft saw that in 80% of investigated incidents, attackers wanted to steal data. A backup on a NAS that is reachable with the same administrator account as everything else does not count as a backup in that scenario.

## What you can do right now

The measures that help in every scenario are surprisingly plain. They are broken down by audience in [how to reduce the chance of becoming a victim](https://cyber.llmnet.nl/en/bescherm-jezelf):

- Know which devices, services, agents and keys you have and which are connected to the internet.

- Use passkeys or a hardware key; get rid of permanent access tokens.

- Patch actively exploited vulnerabilities within a set timeframe (the [CISA KEV list](https://www.cisa.gov/known-exploited-vulnerabilities-catalog) is your priority list).

- Run untrusted code and agents in isolation, without giving them access to your whole home folder or NAS.

- Keep an offline or immutable backup and actually test the restore.

- Let a human decide where an agent can touch money, data or irreversible actions.

If you want to track the threat level yourself, you’ll find the best independent sources in [where you can follow threats and incidents](https://cyber.llmnet.nl/en/dreigingsniveau-volgen). And if you want to see how this looks in practice: read the [most notable AI incidents](https://cyber.llmnet.nl/en/incidenten) like a thriller, but they really happened.

## Sources

- NCSC (National Cyber Security Centre) / NCTV, [Cybersecuritybeeld 2026: AI accelerates and amplifies the threat](https://www.ncsc.nl/nieuws/cybersecuritybeeld-2026-ai-versnelt-en-vergroot-de-dreiging) (6 November 2026)

- Verizon, [2026 Data Breach Investigations Report](https://www.verizon.com/about/news/breach-industry-wide-dbir-finds) (19 May 2026)

- Mandiant / Google Cloud, [M‑Trends 2026 Executive Edition](https://services.google.com/fh/files/misc/m-trends-2026-executive-edition-en.pdf) (March 2026)

- CrowdStrike, [2026 Global Threat Report](https://www.crowdstrike.com/en-us/press-releases/2026-crowdstrike-global-threat-report/) (24 February 2026)

- IBM, [Cost of a Data Breach Report 2025](https://newsroom.ibm.com/2025-07-30-ibm-report-13-of-organizations-reported-breaches-of-ai-models-or-applications,-97-of-which-reported-lacking-proper-ai-access-controls) (30 July 2025)

- ENISA, [Threat Landscape 2026](https://www.enisa.europa.eu/) and Foresight Cybersecurity Threats for 2030

- OWASP GenAI Security Project, [Top 10 for Agentic Applications 2026](https://genai.owasp.org/)

- European Commission, [Safer and more secure digital products](https://commission.europa.eu/news-and-media/news/safer-and-more-secure-digital-products-2026-09-11_en) (11 September 2026)

- Dutch Government, [Cybersecurity Act and Resilience of Critical Entities Act in force](https://www.rijksoverheid.nl/actueel/nieuws/2026/08/15/cyberbeveiligingswet-en-wet-weerbaarheid-kritieke-entiteiten-vanaf-vandaag-van-kracht) (15 August 2026)

- NIST, [Guidelines for transition to post‑quantum cryptography](https://www.nist.gov/news-events/news/2026/10/nist-report-provides-guidelines-transition-post-quantum-cryptography) (27 October 2026)