Rough Seas episode 2: what AI changes in the 2026 threat landscape
In the second episode of Rough Seas, host Ruby Whitmore discusses our Global Threat Landscape 2026 with Sander van der Maden, senior strategic threat analyst and lead author of the report, and Christiaan Ottow, our CTO. Both have worked in cybersecurity for more than 20 years. Their conclusion: AI does not fundamentally change the nature of attacks, but it increases their scale and speed. That calls for a defence that detects and decides as fast as the attacker.
The threat landscape changed faster than ever over the past year. For organisations without their own threat intelligence team, that makes it hard to tell which developments really matter.
AI amplifies and accelerates, but the fundamentals hold
Of all the forces shaping the threat landscape, AI has the most profound impact. Geopolitics, regulation and supply chain vulnerabilities still play a role, but AI amplifies and accelerates each of these developments.
Even so, AI does not change the nature of attacks. Attackers largely use the same techniques, only more often and faster. For defenders, that mainly means applying existing measures faster and more often, such as shorter patch cycles and continuous checks instead of spot checks. What is new is connecting information about your organisation so you can respond to it automatically.
Attackers are ahead of the patch
For a long time, organisations had a reasonably comfortable window between the disclosure of a vulnerability and the first exploitation attempts. That window no longer exists. Google and Mandiant calculated that attackers exploit a vulnerability on average seven days before a patch becomes available. At the same time, half of all organisations take longer than 43 days to install a new patch. In April 2026, less than 13 hours passed between the publication of a vulnerability and its first exploitation.
The cause lies with insecure software, not with AI. AI only makes that problem painfully visible. Patching does remain essential: the average has a wide spread, and attackers exploit many vulnerabilities only after a patch exists.
The next step is focus. The year 2025 brought more than 48,000 new vulnerabilities, but attackers actively exploit only around 1.5 percent of all vulnerabilities. So map your external attack surface, reduce it where possible, and close vulnerabilities in internet-facing systems as quickly as possible. A structured vulnerability management process makes that pace achievable. If an attacker still gets in, segmentation at function and privilege level slows them down. That delay only pays off if you pair it with detection, so you use the time you gain to make decisions.
MFA alone is no longer enough
Identity-based attacks rank among the most prominent developments of the past year. Attackers use stolen credentials, session tokens and access keys to log in as legitimate users to applications such as Microsoft 365 or a CRM platform. AI-assisted phishing kits cost from 30 to 40 euros a month on the dark web. With them, even less skilled attackers steal entire sessions and bypass multi-factor authentication (MFA).
Every organisation our CERT supported in a business email compromise (BEC) incident last year had MFA in place. In 2025, BEC accounted for more than 40 percent of all incidents we responded to, more than cyber extortion. AI-generated phishing also outperforms traditional phishing by more than 50 percent and increasingly appears in flawless Dutch, German and Nordic languages.
The fact that MFA falls short does not mean you can let it go. It means the type of MFA matters. In an adversary-in-the-middle attack, the method we encounter most often, the victim completes the MFA step on a spoofed login page. The attacker then intercepts the session token, which often remains valid after a password reset. Phishing-resistant MFA, such as FIDO2 or passkeys, binds the login to the real domain and makes that interception useless. Start the rollout with administrators and other highly privileged accounts, because that is where an attacker can do the most damage.
Phishing-resistant MFA does not close every route. Attackers also trick employees into giving a malicious app access to company data, or keep using a session they stole earlier. So let only administrators approve apps, and make sure you can revoke sessions and tokens quickly.
Data extortion: the impact starts before you know it
Cyber extortion is shifting away from encryption and towards stolen data. Some groups skip encryption entirely and go straight to databases, for example in a CRM platform. Sometimes the victim's name appears on a leak site, along with the first data samples, before the organisation even knows about the breach. That pushes the organisation straight into the impact phase, with reputational damage, legal consequences and questions from regulators. After all, the stolen data belongs to customers too.
The figures confirm this shift. In 2025, a record number of more than 8,000 organisations worldwide appeared on leak sites. Around 60 percent of victims in our incidents did not pay a ransom. Those who did pay mainly wanted to prevent their data from being published. The attack on Dutch telecom provider Odido in early 2026 showed what this looks like in practice. Criminals used phishing and MFA bypass to access its customer contact platform and stole the data of around 6.2 million customers. Odido refused to pay, after which the attackers published the data in full.
Cloud and SaaS applications such as CRM systems also sit in a layer of the IT stack that organisations typically monitor less closely. A good managed detection and response service therefore looks beyond endpoints and includes signals from identities, cloud environments and applications.
Monitoring helps detect an attack sooner, but it does not limit what an attacker can do with stolen data. Two other measures do. First, an attacker cannot use data you no longer hold as leverage, so delete what neither legislation nor your own retention policy requires you to keep. Second, prepare your communication, because in a data extortion case an incident becomes public within minutes. Agree on core messages and decision lines well before a crisis, along with the values that should guide communication under pressure.
The first AI-driven attack in practice
Agentic AI currently gives attackers the upper hand. An attacker needs to find one weak spot, while the defender has to watch the entire attack surface and weigh every intervention against business operations. AI does help defenders, but only when applied to specific tasks, such as analysing alerts and taking the first steps to contain an attack.
In April 2026, our incident response team responded to an extortion attack that AI systems carried out largely on their own. The attacker entered through two coordinated routes: a vulnerability in a public-facing application and identity weaknesses in the cloud, including an administrator account. What followed was simultaneous activity across Linux systems, applications and databases, without pauses. The attacker pulled data directly from databases and systematically collected API keys and tokens. The pace and variety even led the victim to suspect multiple attackers.
Still, the AI made mistakes, and the defence detected the attack before encryption began. The attack also took an entire weekend, while manual attacks sometimes move faster. In 2025, we handled a heavily automated attack that went from initial access to impact in 22 minutes. There is no reason for panic yet, but there is a reason to adjust defences now.
That adjustment starts with detection. Data from our SOC shows that 64 percent of confirmed threats in 2025 started as medium-severity alerts. High-severity alerts made up 5 percent of the total and accounted for 19 percent of confirmed threats. Organisations that only follow up on the most urgent alerts therefore miss the majority. Against fast, AI-driven attacks, the ability to interpret early, less conspicuous signals in context decides the outcome.
Blurring lines make attribution harder
The boundaries between cybercriminals, nation-state actors and hacktivists are blurring. Ransomware groups collaborate with intelligence services, states use criminal ransomware to maintain plausible deniability, and hacktivism in Europe now aligns almost entirely with nation-state objectives. For a long time, mainly well-resourced states could discover and exploit zero-day vulnerabilities. The latest AI models bring that within reach of cybercriminals. The gap will not disappear entirely, because the effect of AI depends on the attacker's skill and their budget for computing power.
For defenders, this has a practical consequence. The attack method says less and less about the perpetrator and their motive. Base your response on the impact on your organisation, not on the question of who is behind the attack.
Where to start
Despite the pace of change, the fundamentals of defence still hold. We advise organisations to focus on four things:
- Prepare your security programme for a higher pace of attack. Consider three scenarios: attack volumes triple, attacks move four times as fast, and attackers actively use zero-days. Which activities that you currently perform periodically would need to become continuous? Where could you respond faster, and where could you slow an attacker down?
- Understand how your critical assets relate to each other. Maintain an up-to-date view of the relationships between data, systems and identities. When an alert is triggered around an identity, you need to be able to quickly determine which data and systems it could access.
- Make incident response routine, not improvisation. During an incident, only the facts should be new. Roles, meeting structures, communication lines and decision-making should already be familiar through regular training and exercises, for example through cyber crisis readiness exercises.
- Don’t just add controls. Remove what no longer works. Phishing-resistant MFA, conditional access and managed EDR now belong in the baseline. At the same time, regularly review existing measures: are they still effective and do they still add value? Outdated controls can create additional maintenance and exceptions while providing less and less protection.
Also use AI to speed up your own detection and response. Stay level-headed: technology changes fast, but a resilient organisation still rests on the same fundamentals. If you keep those in order, you have the time to adapt.
Listen to the full episode
In the Rough Seas episode, you will hear the full conversation between Ruby Whitmore, Sander van der Maden and Christiaan Ottow, including their account of the AI-driven attack. The second episode of Rough Seas is now available on podcast platforms and on our website.
Want to know more? The full Global Threat Landscape 2026 also covers insider threats, nation-state attacks on commercial networks and hacktivism, with concrete measures for each threat: Northwave Global Threat Landscape 2026.
Sources:
We are here for you
Need help with your cyber security or wondering how secure your business really is?
Get in touch and we will help you find the solution you need.