Analysis Reveals How Multi-Agent Tools Enable Less-Skilled Actors to Execute Complex Attacks
According to a recent threat report released by Anthropic, a leading figure in AI technology, the dynamics of malicious cyber activities have shifted significantly. It has become apparent that threat actors no longer require advanced skills or resources to execute sophisticated attacks, a paradigm shift facilitated by artificial intelligence.
Anthropic’s Threat Intelligence team conducted an extensive investigation into various attempted cyber incursions between December 2025 and August 2026. Interestingly, their focus was not just on the rapidity with which AI systems can generate exploits but on the broader and more intricate attacks that can be orchestrated with limited resources. This new approach allows even less-skilled individuals to engage in complex operations that would have traditionally required significant technical know-how.
The company pointed out that “sophisticated and persistent threat actors continuously test our safeguards and try to circumvent the technical measures we use to detect and prevent misuse.” Anthropic documented several operations where malicious actors utilized their more accessible models—Haiku, Sonnet, and Opus—rather than the more advanced Claude Mythos or Fable models. The variety of attacks identified spanned numerous sectors, including cyber operations, influence campaigns, surveillance, scams, and even biological misuse.
Anthropic noted that many of these operations were empowered by AI through direct execution or orchestration. The company highlighted that the use of AI transcended basic interactions typical of chatbots, instead involving “multi-agent frameworks” capable of executing reconnaissance, exploitation, and data exfiltration tasks. Furthermore, Anthropic successfully disrupted multiple actions by threat actors from various locations, including Russia and the cybercriminal group ShinyHunters.
Among the notable incidents was a case involving a Russian-speaking operator named “JackPoterz,” who targeted military intelligence agencies within Ukraine and European governments. This particular operation employed custom AI workflows designed to automate crucial processes such as infrastructure acquisition, phishing campaigns, and data extraction. Another Russian-speaking individual was reported to have exfiltrated approximately 26 gigabytes of sensitive information from hotel and financial technology platforms while employing malicious prompt injections in sandbox environments.
Interestingly, even a small group of “low-level” hacktivists managed to enhance their capabilities significantly through AI integration. Anthropic observed these activities during the spring of 2026 when a French-speaking user began directing their efforts toward European political parties, media outlets, think tanks, and their corresponding SaaS providers. This group developed custom Rust-based scanners that deployed Claude’s coding abilities to identify exposed API keys, showcasing the democratization of sophisticated hacking techniques.
Additionally, members of the ShinyHunters collective employed Claude models to analyze APIs, facilitate bulk data exports, and create and modify tokens, further exemplifying the breadth of malicious uses for AI technology. Moreover, Anthropic’s findings raise serious concerns about the implications of AI-enhanced capabilities for cyber defense strategies. The company announced that such advancements threaten to “quickly and easily subvert” the heightened costs defenders typically incur to combat adversaries, as traditional approaches now face greater challenges in detection and mitigation.
Moreover, while hacktivist groups and politically motivated players comprise a portion of AI’s malicious applications, Anthropic also scrutinized instances where AI models were attempted for more dangerous uses. The report unveiled efforts to build biological threats, although Anthropic maintained that the specifics surrounding these attempts were redacted based on caution. Researchers from a military-affiliated institute reached out for assistance in writing grant applications concerning the chikungunya virus, prompting Anthropic to reject the inquiry due to the questionable nature of the request and the institutional ties involved.
Another group focused on bird flu adaptations found themselves accessing Claude through restricted regions via VPNs. Anthropic’s revelation illustrates not only the escalating scale of cyber threats but also emphasizes how the use of AI has enabled less-skilled actors to carry out intricate attacks in a markedly efficient manner.
This trend has not gone unnoticed by other cyber defense teams. Palo Alto Networks’ Unit 42 reported that cyberattacks are becoming more rapid and complex and that organizations should consider utilizing more AI agents in their cyber defense strategies. They echoed sentiments made by OpenAI’s president, Greg Brockman, regarding the narrowing defense window due to evolving attack techniques.
Furthermore, the Google Threat Intelligence Group noted that adversaries are increasingly employing multi-agent frameworks to manage operational pipelines and credential harvesting. These developments point to a concerning future in which malicious activities proliferate at an alarming rate, driven in part by the democratization of AI technologies that enable a new era of cyber threats.