AI-powered cyberattacks in 2026: An overview of the new threat landscape

22 seconds: that was how long it took in 2025 for initial access groups to pass on compromised access to the next group of attackers, such as ransomware operators. In 2022, this figure was still over eight hours (Mandiant, M-Trends 2026). The figure illustrates the direction in which the threat landscape is moving: attacks are carried out through a division of labour and are automated, with AI further accelerating many of the steps involved. Anyone who only reacts hours after the initial breach will, in many cases, be too late.

AI makes attackers faster, but not yet invincible

The Google Threat Intelligence Group has observed that state-sponsored groups from China, Iran and North Korea, amongst others, are using generative AI for target research, credible pretexts, vulnerability analysis and code development. However, none of these actors has yet achieved capabilities that fundamentally alter the threat landscape (GTIG AI Threat Tracker, February 2026). Mandiant takes a similar view of the year 2025: most successful attacks continued to rely on human and systemic weaknesses.

A new development is malware that utilises AI directly during an attack. Malware families such as PROMPTFLUX and PROMPTSTEAL query language models whilst running in order to evade detection. In May 2026, the BSI spoke of a “new era of cyber security” in reference to AI models capable of independently identifying security vulnerabilities and generating appropriate attacks. The time window between the discovery and exploitation of a vulnerability is shrinking (BSI, May 2026).

German businesses are already feeling the impact: the proportion of companies suffering losses due to deepfakes has doubled from 4% to 8%, whilst losses caused by automated robo-calls have risen from 3% to 14% (Bitkom, Economic Protection 2026).

Damages are falling, but the pressure on businesses remains high

According to Bitkom, the German economy suffered losses of at least 211 billion euros in 2026 as a result of data theft, espionage and sabotage, down from 289.2 billion euros the previous year. Around three-quarters of this was attributable to cyber-attacks. This is no cause for complacency: 96 per cent of companies were affected or suspect an incident, whilst two-thirds expect more attacks in the coming twelve months. Only 43 per cent now consider themselves very well prepared, down from 50 per cent the previous year.

The BSI assesses the situation as “tense” and identifies inadequately protected attack surfaces in particular as the cause of Germany’s vulnerability (BSI Situation Report 2025).

Threat landscape in figures

• 119 new vulnerabilities per day, an increase of around 24 per cent (BSI)

• 32 per cent of all attacks investigated began with an exploit (Mandiant)

• 14 days was the median time attackers remained undetected, compared with 11 days the previous year (Mandiant)

• 96 per cent of companies in Germany were affected or suspect an incident (Bitkom)

• 82 per cent expect attackers to make greater use of AI (Bitkom)

Many companies are still investing in security that looks good, and not enough in security that actually works. Attackers have long been using AI to become faster. A security strategy that takes a break at night and at weekends is no answer to this. What matters is not how many tools are in use, but how quickly someone reacts when things get serious.

Andreas Papadaniil, CEO of suresecure

Ransomware remains the most profitable business model

According to the BSI, professionally organised extortion groups remain the greatest threat. Ransomware-as-a-Service remains at a high level; the number of ransomware attacks reported to the BKA stood at 950, virtually unchanged. Around 80 per cent of these targeted small and medium-sized enterprises.

Technically, the approach has shifted. Many attacks combine encryption with data theft and the threat to publish the data. Backups remain indispensable, but do not help against data leaks. Added to this is the division of labour: According to Mandiant, in ransomware operations, a pre-existing compromise was the most common point of entry in 2025, accounting for 30 per cent of cases; a year earlier, the figure was 15 per cent. Failing to detect an initial breach often opens the door to the next group of attackers.

Automation is shifting the race to exploit vulnerabilities

Exploits have been the most common point of entry into corporate networks for many years. Mandiant now estimates the average time to exploitation of a vulnerability at minus seven days: attackers therefore regularly exploit vulnerabilities before a patch is available.

This is made possible by scripts and AI-powered tools that attackers use

This lowers the barrier to entry for less experienced attackers, whilst professional groups increase their output. Under these conditions, patch cycles lasting several weeks are no longer sufficient.

Geopolitical tensions are affecting small and medium-sized enterprises

State actors are appearing significantly more frequently in German companies. According to Bitkom, 37 per cent of affected companies were able to attribute at least one incident to foreign intelligence services; in 2025 the figure was 28 per cent, and in 2023 it was still 7 per cent. 52 per cent attributed at least one attack to China, and 49 per cent to Russia.

This means that even sectors which do not regard themselves as strategic targets are now being targeted – for example, as suppliers to critical infrastructure or as gateways to larger partners in the supply chain.

The attack surface is growing faster than security measures

Cloud services, hybrid working models, IoT devices and close links with partners and service providers are constantly creating new points of entry. The BSI has observed that attackers specifically target the least well-protected attack surfaces. Web applications, in particular, are frequently inadequately secured.

Combined with automated scans, this increases the likelihood that attackers will find a vulnerability before the organisation’s own team is aware of it. It is therefore crucial how quickly a company detects unusual activity. After all, according to Mandiant, organisations had already detected 52 per cent of attacks themselves by 2025, up from 43 per cent the previous year.

Basic measures and a rapid response help against AI-powered attacks

AI changes the pace of attacks, not their underlying logic. The BSI therefore recommends first and foremost consistently implementing the basic measures, as these are also effective against AI-enabled attacks. This results in five priorities:

• Harden systems: Install security updates promptly, configure systems securely and segment networks.

• Protect identities: Limit access to what is necessary and implement multi-factor authentication across the board.

• Establish visibility: Centrally collect and analyse security-relevant data from endpoints, the network, the cloud and identities, for example via SIEM and EDR.

• Respond around the clock: A Security Operations Centre, whether in-house or as a managed service, detects suspicious activity and initiates countermeasures, even at night and at weekends. Automated processes (SOAR) reduce the time to containment.

• Preparing people: Staff are aware of current deception tactics such as deepfakes and calls using synthetic voices; payment and authorisation processes include mandatory verification via a second channel.

Added to this is the defensive aspect of AI: the BSI recommends actively using AI tools for vulnerability detection. Technology alone is not enough. It only becomes effective when combined with clear processes and a team capable of assessing alerts and making decisions.

FAQ