Artificial intelligence safety company Anthropic released a threat intelligence report detailing how malicious users are attempting to exploit its Claude AI model for biological weapon research and cyberattacks, while separate findings from Microsoft reveal threat actors simultaneously weaponizing AI brand names as phishing lures. According to Anthropic, the dual-use nature of advanced scientific knowledge means capabilities designed to help researchers find medical cures can also be twisted toward dangerous applications.
Biological Research Risks Documented by Anthropic
Anthropic’s threat report highlights five specific case studies involving threat actors attempting to use Claude for biological threat creation. According to the company, one case involved a user researching methods to make a mosquito-borne disease more transmissible. Another instance focused on redesigning toxins for what Anthropic identified as a national program. These findings underscore a central dilemma for policymakers and AI developers: the same technical information required to understand pathogens and develop disease cures can potentially be adapted to engineer biological weapons.
Cybercriminal Exploitation and Vibe Hacking Tactics
Beyond biological inquiries, cybercriminals are systematically weaponizing AI models for sophisticated digital assaults. In one campaign tracked as GTG-2002, threat actors used the Claude Code platform to execute large-scale data extortion across at least 17 organizations in a single month. This setup allows the AI to autonomously scan endpoints, manage credential sets, develop obfuscated malware, and draft targeted extortion notes exceeding $500,000.
Did you know?
Threat actors have configured AI platforms to make autonomous strategic decisions during active network penetrations, dividing attacks into distinct phases from reconnaissance to psychological extortion, according to security reports.
AI Brands Weaponized in Global Phishing Campaigns
While malicious actors misuse underlying models, they are also exploiting public interest in artificial intelligence as a social engineering tactic. According to Microsoft Threat Intelligence, campaigns impersonating popular AI platforms—including ChatGPT, Microsoft Copilot, DeepSeek, and Anthropic’s Claude—have surged to deliver phishing payloads and malware. These attacks do not represent a compromise of the AI services themselves. Instead, threat actors use trusted branding to lower user skepticism.
Microsoft noted a prominent campaign on May 5, 2026, where attackers sent thousands of ChatGPT-themed emails urging targets to update their subscription payment methods or face account downgrades. These multi-stage redirection chains routed victims through legitimate abused domains to harvest credit card data and credentials across regions including South Africa, Switzerland, and Austria.
Frequently Asked Questions
What security risks does Anthropic’s report highlight regarding Claude?
Anthropic’s report details attempts by threat actors to use Claude for biological weapon research—such as increasing the transmissibility of mosquito-borne diseases—and sophisticated cyber attacks including data extortion and malware development.
Are AI platforms like ChatGPT and Claude actually compromised in phishing campaigns?
No. According to Microsoft Threat Intelligence, threat actors are merely using AI brand names, logos, and hype as social engineering lures in phishing emails and malvertising, rather than compromising the actual vendor infrastructure.
What is “vibe hacking”?
Stay Informed on AI Security
Subscribe to our newsletter for the latest updates on threat intelligence, artificial intelligence safety research, and cybersecurity trends.
Keep reading