New threat landscape for companies
AI tools have been used for cyberattacks for years, such as for automated phishing campaigns or the generation of malicious code. In April 2026, the US-based AI company Anthropic announced the launch of Anthropic AI for Enterprise. Presentation of Claude Mythos Preview, a new general language model, but clearly indicating that this development has reached a new dimension.
Mythos Preview and Project Glasswing: A New Dimension of Cyber Risks
According to Anthropic, the new model was able to identify thousands of previously unknown security vulnerabilities of high and critical severity within a short period of time and, upon request, exploit them as well. Affected were common operating systems and web browsers. Interestingly, many of these vulnerabilities are extremely subtle and difficult to detect and have existed for decades. These are so-called zero-day vulnerabilities, that is, errors whose existence was previously unknown.
For example, a 27-year-old vulnerability was identified in the OpenBSD operating system, which is considered particularly secure and is used in, among other places, critical infrastructure. According to Anthropic, even individuals without formal security training were able to develop functional exploits with the help of the model.
Anthropic deliberately did not make the model publicly available and instead published the Project Glasswing It is a collaboration with security researchers, companies, and authorities with the goal of identifying vulnerabilities in common software and IT infrastructure through the use of the new model and responsibly closing them before models with comparable capabilities become widely available and can be misused.
Nevertheless, it is foreseeable that models with comparable capabilities will become more widely available in the future and can therefore be used for abusive purposes as well. Companies should incorporate this development into their risk assessment early on.
How the threat landscape is changing
1. Decreasing entry barriers
Activities such as vulnerability analysis, exploit development, or circumventing security systems have previously required specialized expertise and significant resources. By using AI, these processes can increasingly be automated. This significantly expands the range of potential attackers, as less specialized actors can also access corresponding tools.
2. Acceleration of attacks and shortened reaction times
The tests described by Anthropic not only reveal new attack vectors, but above all a drastic acceleration of existing processes. For example, Mythos Preview was able to develop functional exploits within a few hours – a process that previously could take days or weeks even for experienced security experts.
The time between the publication of a vulnerability and its actual exploitation („Time-to-Exploit“) is significantly shortened. Security vulnerabilities must therefore be detected and closed much faster than before. Delays that were previously tolerable could in the future lead directly to real attacks.
3. Attacks on unknown vulnerabilities
Classical protection mechanisms such as virus scanners, firewalls, or intrusion detection systems regularly rely on known attack patterns. They reach their limits when attacks are based on previously unknown vulnerabilities. Companies should therefore align their IT architecture according to the principle of multi-layered security, so that even if a single vulnerability is exploited, a complete compromise is not achieved.
4. Scaling and automation of attacks
AI enables parallel and continuously optimized attacks on a wide variety of systems with minimal additional effort per target. Current reports show a significant increase in attacks by actors using AI. This fundamentally changes the structure of cyberattacks. They are becoming broader, faster, and more systematic.
5. Increasing pressure for automation on the defense side
At the same time, it becomes clear that AI models already available today are capable of identifying vulnerabilities on a large scale. Companies can actively take advantage of this development, for example for automated software analysis, prioritizing security notifications, or assisting in the remediation of vulnerabilities.
Furthermore, the pressure to automate existing processes is increasing. The expected increase in vulnerability reports and attack attempts will no longer be manageable solely manually. In particular, the response to security incidents, the analysis of alerts, and the prioritization of measures must be supported more by technical systems, including AI.
Importance for companies
For companies, the requirements for IT security are rising significantly. The drivers are both the rapid development of AI and increasing regulatory requirements. Article 32 of the General Data Protection Regulation (GDPR) requires the implementation of technical and organizational measures in line with current best practices. This is continuously evolving through AI. The NIS2 Directive and its national implementation oblige affected companies to comprehensive risk management, technical protection measures, and reporting obligations; the responsibility lies explicitly with the management. With the European AI Act, additional requirements for transparency, security, and risk management are added for certain AI applications.
Conclusion and recommendations for action
Current developments show that the technical prerequisites for cyberattacks are fundamentally changing. Artificial intelligence acts as a catalyst and amplifier of existing risks. Therefore, companies should regularly check whether their technical and organizational measures still comply with current standards and, if necessary, adapt them.
An early and structured approach to addressing the new risks is crucial to avoid security gaps and meet regulatory requirements.
1. Short-term measures
2. Medium-term measures
3. Strategic measures
Check the mandatory fields –
Step-by-step in the webinar
If you would like to be there or to ask your questions:
Get in touch – the registration details will follow shortly.
FAQ
How has artificial intelligence changed the threat landscape for companies?
AI can already be used today for automated phishing campaigns, the generation of malicious code, and the analysis of security vulnerabilities. According to the article, current models are now reaching capabilities that can also identify and exploit previously unknown and hard-to-detect vulnerabilities. 2026-05_04_Artificial_Intelligence…
Why does AI reduce the entry barriers for cyberattacks?
Activities such as vulnerability analysis, exploit development, or circumventing security systems have previously required specialized expertise and significant resources. AI can automate parts of these processes, thereby also opening up corresponding attack possibilities to less specialized actors.
Why must companies close security gaps faster in the future?
AI can significantly accelerate the development and execution of attacks. This reduces the time between the discovery of a vulnerability and its actual exploitation. Patch and response processes therefore need to act faster than before.
Why are traditional protective mechanisms no longer enough?
Virus scanners, firewalls, and intrusion detection systems often rely on known attack patterns. When attacks target previously unknown vulnerabilities, such mechanisms reach their limits. The article therefore recommends a multi-layered security architecture in which the exploitation of a single vulnerability does not directly lead to complete compromise.
What legal requirements are relevant for cyber risks caused by AI?
The article specifically refers to Article 32 of the GDPR, the NIS2 requirements, and the AI Act. Companies must align their technical and organizational measures with the state of the art, systematically manage risks, and – depending on the specific case – consider additional requirements for security, transparency, and risk management.
What measures should companies take now?
In the short term, in particular, patch cycles should be reviewed, externally accessible systems should be fully captured, and incident response plans should be updated. In the medium term, the article recommends, among other things, AI-based security solutions, clear internal AI policies, and regular employee training. Strategically, information security management should be further developed and compliance with regulatory requirements should be systematically documented.
