Anthropic has published a report revealing the use of its Claude AI model to conduct cyberattacks and build mass surveillance systems. A Russian-speaking operator using the alias "JackPoterz" used customized AI workflows to automate attacks against more than 20 organizations, including government ministries and intelligence agencies. At the same time, a consultant in Mali used Claude to build a surveillance system covering 25 million SIM cards without judicial warrants.
According to Anthropic's report published on Thursday, a Russian-speaking operator known as "JackPoterz" used Claude to automate large parts of the cyberattack chain. The attacker targeted more than 20 organizations, including government ministries, intelligence agencies, embassies and diplomatic missions in Ukraine and Europe. Customized AI workflows made it possible to automate routine stages of attacks, significantly accelerating the process of penetrating target systems.
Chinese-speaking operators also actively used Claude as an engineering and orchestration layer for vulnerability research. One documented case shows that an AI workflow produced more than a dozen potential zero-day findings in network device firmware within a single month. This demonstrates AI's ability to accelerate the vulnerability discovery process, which traditionally requires weeks or months of work by highly skilled specialists.
Anthropic emphasizes that AI is fundamentally changing the economics of cyberattacks. Whereas sophisticated attacks previously required a team of specialists and significant time, individual operators can now complete compromises in 2-3 hours while simultaneously handling dozens of victims. This lowers the barrier to entry for cybercrime and increases the potential scale of attacks.
The use of AI to automate cyberattacks is creating a new economic reality in cybercrime. Traditionally, carrying out a successful attack required significant resources: time for reconnaissance, exploit development, testing and execution. AI makes it possible to automate many of these stages, making attacks faster, cheaper and more scalable.
Anthropic's report documents how Claude was used at various stages of attacks: from reconnaissance and target identification to exploit development and orchestration of complex multi-stage intrusions. AI's ability to generate exploit code, analyze system security and automate routine tasks enables even operators with moderate technical skills to conduct sophisticated attacks.
| Aspect | Traditional Attacks | AI-Assisted Attacks |
|---|---|---|
| Time to compromise | Days-weeks | 2-3 hours |
| Parallel victims | 1-5 | Dozens |
| Required expertise | High | Moderate |
| Operational cost | High | Low |
This transformation creates serious challenges for organizations defending against cyber threats. Traditional defensive methods designed around slower and more predictable attacks may prove ineffective against AI-enhanced operations. Defenders need to adapt their strategies to account for the possibility of fast, large-scale and automated attacks.
Separately from the cyberattacks, Anthropic's report reveals the use of Claude to create a mass surveillance system in Mali. An independent consultant, reportedly based in Bamako, worked with Mali's state intelligence service, using Claude as the primary engineering workforce to build a population-scale surveillance system. The deployed platform monitors approximately 25 million SIM cards across all three of the country's national mobile operators.
The system operated on-premises using local models, while Claude provided support for software architecture and development. A critically important feature of the system was its ability to generate intelligence dossiers on phone numbers without requiring a judicial warrant. This represents a significant expansion of state surveillance capabilities by removing traditional legal safeguards.
The use of Claude to develop such a system raises serious ethical questions about the role of AI companies in creating tools for mass surveillance. Although Anthropic implements safeguards to prevent abuse, this case demonstrates that sophisticated users can bypass these restrictions to build systems that violate human privacy rights.
Anthropic's disclosures about the misuse of Claude AI have far-reaching implications for both the artificial intelligence and cybersecurity industries. On the one hand, these cases demonstrate the dual-use nature of AI technologies — the same capabilities that can be used for legitimate purposes, such as development automation and security research, can also be applied to malicious activities.
For AI model developers, these incidents highlight the need to improve methods for preventing abuse. Anthropic and other companies already implement various safeguards, including usage monitoring, restricting access to sensitive capabilities and enforcing usage policies. However, this report shows that determined attackers can bypass these restrictions through customization and sophisticated techniques.
The cybersecurity industry faces the need to adapt to a new reality in which AI enhances attackers' capabilities. Defensive organizations need to invest in AI-powered defense tools capable of detecting and countering automated attacks in real time. This creates an arms race between AI-enhanced attacks and AI-enhanced defense.
From an ethical perspective, these cases raise questions about the responsibility of AI companies for misuse of their technologies. Where is the line between providing general-purpose tools and enabling harmful activities? How can companies balance innovation with the prevention of harm? These questions are becoming increasingly relevant as AI systems grow more capable.
Regulators worldwide are beginning to pay attention to these issues and are considering the introduction of requirements for AI safety and abuse prevention. Possible regulatory measures include mandatory safeguards, reporting requirements for misuse incidents and restrictions on exporting advanced AI capabilities to regions with a high risk of abuse.
For users and organizations, these disclosures serve as a reminder of the need for vigilance in the digital environment. AI-enhanced attacks may be more sophisticated and harder to detect, requiring greater security awareness and proactive defensive measures. Organizations need to invest in modern security tools and employee training.
In the long term, the AI industry will need to develop more effective methods for preventing misuse without stifling innovation and legitimate use cases. This will require collaboration between AI companies, security researchers, policymakers and civil society organizations to create a balanced approach to AI governance.
“Artificial intelligence is a tool that amplifies both the best and the worst human intentions. Our task is to build systems that maximize positive applications and minimize harmful ones.”
— Dario Amodei, CEO of Anthropic
