Every day, companies in Germany process invoices, contracts, personnel files, and customer data. An increasing portion of this work is handled by AI systems. As the total number of systems, access points, and people continues to rise, so does the attack surface through which an attacker can gain access to the company.
For the study „Economic Security 2025“ on corporate security, the Digital Association commissioned Bitkom From April through June 2025, a total of 1,002 companies with ten or more employees were surveyed by telephone. The results: 87 percent had been affected by data theft, industrial espionage, or sabotage in the past twelve months, while another 10 percent suspected they had been. The total damage amounted to 289.2 billion euros, 70 percent of which was attributable to cyberattacks.
Artificial intelligence is changing the landscape for both sides. Attackers craft error-free phishing emails, clone voices, and find vulnerabilities more quickly. Defenders use AI to detect unusual activity on their own networks. Two-thirds of the companies surveyed (66 percent) believe that AI is being used more frequently in attacks against them. At the same time, only six percent are using AI to improve their own IT security.
This article describes how the threat landscape has shifted and what companies should look out for when it comes to technology, data, and providers. It also explains why training employees has become a legal requirement in the age of AI. The article provides an overview of the tools available on the defense side in Cybersecurity in the Age of AI.
The Threat Landscape in the Age of AI
More Vulnerabilities, Smaller Targets
The Federal Office for Information Security (BSI) describes in its 2025 Management Report For the period from July 2024 to June 2025, the situation is expected to „remain at a high level.“ An average of 119 new vulnerabilities in IT systems were reported each day, an increase of about 24 percent compared to the previous reporting period.
Of the attacks reported to the authorities—such as those involving ransomware (extortion software that encrypts data)—about 80 percent targeted small and medium-sized enterprises (SMEs). According to the BSI, these companies often lack the resources and knowledge to protect themselves.
So attackers choose the easiest route, and this often involves systems that lack the latest security updates and human error. The Data Breach Investigations Report 2026 Verizon analyzed security incidents worldwide between November 2024 and October 2025. It identified software vulnerabilities as the most common point of entry, accounting for 31 percent of cases—ahead of stolen credentials—and found that ransomware was involved in 48 percent of all data breaches examined.
AI on both sides
IBM works with the Ponemon Institute to measure the cost of a data breach annually, based on actual data breaches at companies worldwide. In the Cost of a Data Breach Report 2026 The global average cost of a data breach is $4.99 million, up 12 percent from the previous year and a record high.
According to the same report, the number of AI-enabled attacks rose by 56 percent, primarily due to deepfake imitations (AI-generated fake voices or videos) and AI-enabled malware. Verizon identifies 15 attack techniques that are now being amplified by generative AI.
Bitkom President Dr. Ralf Wintergerst commented on this in the Press Release Regarding the Study September 18, 2025: „Artificial intelligence enables entirely new methods of attack, such as deepfakes, but it also enhances traditional attacks, for example through much more credible phishing emails.“ And: „It is important to train employees and raise their awareness of these issues.“
So far, deepfakes have caused damage to only 4 percent of the companies surveyed. However, 11 percent have already reported deepfake attacks that caused no damage.
What Companies Should Keep in Mind Regarding IT and Data Security
The reports from BSI, Bitkom, and Verizon paint a similar picture: Most successful attacks exploit known vulnerabilities, weak access points, or human error. This leads to six key points that apply to companies of all sizes.
- Vulnerability Management: The BSI identifies protecting one's own attack surfaces as the key lever for 2026: timely updates, restrictive access management, and as few publicly accessible systems as possible.
- Accesses Based on the Principle of Least Action: Each person is granted only the permissions required for their role, and every access event is logged. This principle is at the heart of Zero-Trust Architectures, in which every access attempt is verified, even from within the organization's own network.
- Data Storage and Provider Selection: Every procurement process must specify where data is stored, which subcontractors are involved, and what certifications the provider holds, because these factors determine data security. 67 percent of the companies surveyed by Bitkom consider themselves dependent on security solutions from the United States.
- Emergency Plan: According to Bitkom, six out of ten companies have a contingency plan for security incidents, while 39 percent are unprepared. A plan outlines reporting procedures, recovery, and communication before an emergency occurs.
- Third-Party Audit: Penetration tests (simulated attacks on a company's own systems) and code reviews uncover issues that remain hidden internally. This is especially true for solutions that teams develop themselves with AI support.
- Rules for AI Tools: It is necessary to specify which data may be entered into an AI tool, which tools are approved, and who reviews the results. Without these rules, customer data ends up in third-party models without anyone noticing.
Five of the six points can be purchased, set up, and documented. Whether the rules for AI tools are followed, however, is decided every day at every workplace—and thus by the employees.
Why Employee Training Is Crucial
Humans as a Target
Social engineering refers to techniques in which attackers exploit human behavior to gain access or obtain information, for example, through fake emails or deceptive phone calls.
According to Bitkom, nearly one in two companies has experienced such attempts in the past twelve months: 21 percent frequently, and 28 percent occasionally. Phishing attacks caused damage to 22 percent of companies, while password attacks affected 21 percent.
Generative AI reduces the cost of these attacks. A phishing email written in flawless German, in the tone of the recipient’s supervisor, and referencing a real project can now be generated in seconds.
It’s even harder to spot on a smartphone: Verizon reports click-through rates that are 40 percent higher on mobile devices, which is why attackers are switching to text messages and phone calls. Such emails usually slip past the spam filter. At that point, only the employees can stop the attack.
The Training Gap in Numbers
According to the Bitkom study, while the majority of companies do provide regular training on phishing or social engineering, this training is not comprehensive. Fifty-five percent provide training only to selected departments or positions, while 24 percent train all employees. Twenty percent do not provide any IT security training at all.
When it comes to working with AI itself, the gap is even wider. The TÜV Association From January through March 2026, it commissioned Forsa to survey 500 companies with 20 or more employees. Twenty-seven percent had trained their employees in the use of AI, more than twice as many as in 2024 (12 percent). Forty-five percent saw no need for such training.
A Bitkom Survey Of 603 companies surveyed in the summer of 2026, 70 percent are now training their employees in the use of AI. Of these, 37 percentage points are accounted for by companies that train only select employees. The difference between the two figures can be explained by the later survey date, a different group of respondents, and a broad definition of training—ranging from one-time information sessions to programs that include exercises.
Obligations under NIS2 and the AI Regulation
Training is now also a legal requirement. The NIS2 Implementation Act refers to the affected companies as „particularly important and important facilities.“ Their management is required to Section 38 of the BSI Act to participate regularly in training sessions in order to be able to assess cyber risks and the measures to counter them.
Under paragraph 2, management is liable to its institution for damages it causes through its own fault by failing to implement or monitor the risk management measures set forth in paragraph 1.
The AI Regulation applies to everyone who uses AI systems. Article 4 Since the version dated July 27, 2026, it has required providers and operators to take measures to promote their staff’s AI literacy, tailored to their prior knowledge and the context in which they work.
The regulation does not prescribe a specific level of competence, but it also does not permit inaction. The article on the regulation describes what else it requires: EU AI Act for Businesses.
This Is What an Effective Training Program Looks Like
Traditional training: a presentation on passwords and phishing once a year, with attendance confirmed by signature, and the same content for everyone. This format falls short against AI-powered attacks because they rely on timeliness, personalization, and speed. A program designed to counter this follows the attackers’ logic.
- Segmenting target audiences: Management (mandatory training for NIS2 organizations pursuant to Section 38 of the BSIG, liability, decision-making in emergencies), specialized departments (phishing, phone calls, data sharing), and developers (secure use of AI tools in programming).
- Practice current attack patterns: Deepfake calls, fake invoices with altered bank account information, and messages sent in the name of company management.
- Establish guidelines for using AI tools: Work through the rules from the checkpoint above using examples from your own daily work until you have them down.
- Practice reporting procedures: A suspected case must be reportable within one minute, without assigning blame. The sooner a report is made, the less damage is caused.
- Repeat and measure: Short training sessions several times a year, phishing simulations (test phishing emails sent by the company itself) with an analysis by department, and follow-up training where the click-through rate remains high.
Tip: The click-through rate of your own phishing simulation is the most meaningful metric because it measures behavior rather than simply counting participants. Along with the status of updates and the backup test, it should be included in every security report to management.
Example: A development team building an internal application with AI support learns in the Konfuzio Academy a case from our own task list. It concerns the guidelines for AI tools in everyday use: what data can be entered into which tool, who verifies it, and what is logged. Afterward, a penetration test evaluates the running system, and the test report is sent back to the same team.
Security at the Service Provider: The Example of Konfuzio
These requirements apply doubly to document processing providers because they handle their customers’ most sensitive data. Konfuzio is developed by Helm & Nagel GmbH in Germany, operated in accordance with ISO 27001 (the international standard for information security), and hosted in German data centers.
Certificates, security measures, and subcontractors can be viewed in the Trust Center; details are available on the page Security and Compliance. Helm & Nagel also offers the expertise gained from its own operations as a service, ranging from compliance consulting to penetration testing.
Conclusion
In the age of AI, the threat landscape has shifted toward more attacks using known methods.
119 new vulnerabilities every day, 80 percent of reported attacks targeting SMEs, and 87 percent of companies affected point to a clear trend. Attackers seek out poorly protected areas and use AI to make known methods faster and more convincing. Technology and policies remain the foundation of corporate security—from managing attack surfaces and access rights to selecting vendors.
Employees are a vulnerable target that is rarely protected in a systematic way. Only 24 percent of companies provide IT security training to all employees, and when it comes to working with AI, the percentage ranges from 27 to 70 percent, depending on the survey and definition.
At the same time, Section 38 of the BSIG requires the management of NIS2 entities to undergo regular training, and Article 4 of the AI Regulation requires providers and operators to implement measures to ensure AI literacy among all those who operate AI systems. A program that targets specific audiences, addresses current attack patterns, and delivers measurable results is therefore an investment whose effectiveness can be gauged by the click-through rate after each session.
Would you like to know where your IT infrastructure and teams stand—from vendor evaluation to penetration testing? Simply send a message to Konfuzio to get expert advice.
EN 
DE 
ES 
FR