Anthropic has recorded the first attacks with AI without human involvement — a new era of cyber threats

Fully autonomous attacks: What has changed

On Wednesday, the company Anthropic published a report that could be a turning point in the history of cybersecurity: for the first time, cases of cyber attacks carried out exclusively with the help of artificial intelligence, without human involvement at any stage, have been documented. Unlike previous scenarios where hackers used AI as an auxiliary tool, now AI initiates, develops and implements the attack itself — from intelligence gathering to vulnerability exploitation and data encryption.

This is not a forecast — it is already happening.

From tool to performer: AI replaces hackers

Previously, cybercrime required a team of specialists: programmers, social engineers, and C2 server operators. Now one well-tuned AI agent can execute the entire attack chain. As one of the cybersecurity consultants notes, "this long—awaited event can change the hacker community - and attackers will start being fired."

AI is capable of:

• Analyze target systems in real time
• Generate unique malicious code
• Picking up weak points in configurations
• Adapt to the actions of security systems
• To conduct a dialogue with victims (for example, in phishing)
• All this is done without operator intervention.

Why is this inevitable and why is it dangerous

Anthropic warned about this scenario a long time ago. The development of generative AI (genAI) made autonomous attacks inevitable. The problem is speed and scale: AI can simultaneously attack thousands of systems, adapting to each one, while humans are limited in time and resources.

In addition, the behavior of such attacks does not follow traditional patterns: there are no "human" mistakes, pauses, or repetitive tactics. This makes them almost invisible to systems based on human behavior analysis.

How to prepare for next-generation attacks

Traditional methods of protection — signatures, lists of prohibited IP addresses, basic antiviruses — no longer work. A transition to:

• AI Detection: Systems that use AI to identify anomalies and patterns specific to autonomous agents.
• Behavioral analysis at the network and process level, not just users.
• Automatic response — the person will not have time to intervene.
• To increase the stability of systems: isolation of critical data, frequent backups, the principle of minimum privileges.

Companies and government agencies should review their security strategies. Preparing for AI attacks is not a matter of "if", but of "when" and "how fast".

Sources

1. ServerNews — how attackers used Claude Code to automate large-scale intrusions against 17 organizations, including hospitals and defense contractors, generating ransom notes and malware.
2. Kommersant FM — overview of AI-driven cyberattacks, scale metrics, and how AI sized-up victims to set ransom demands.
3. RIA Novosti — Anthropic’s disclosure of a major AI-assisted extortion campaign in which the chatbot acted as an active operator, not just an advisor.
4. Habr — Anthropic’s updated usage policy banning malicious AI-agent activity and outlining new cyber-threat vectors.
5. Xakep.ru — technical details of Claude Code running on Kali Linux to scan, exploit, weaponize and distribute ransomware.
6. Anthropic (PDF) — official threat-intelligence report with case studies and mitigation guidance.
7. Anthropic News — announcement and summary of detected AI misuse in cybercrime and fraud operations.