Trust as a Vulnerability: How Legitimate Domains Became Hacker Tools
Corporate security systems are conditioned to trust big tech, and this blind faith has turned openai.com and anthropic.com into the perfect environments for malware delivery. According to research from Push Security, cybercriminals have begun mass-exploiting "shared chat" features to bypass filters and antivirus software. The mechanics are as simple as they are cynical: attackers purchase search engine ads that lead to a legitimate chat page, which standard security tools wave through without a second thought.
The Anatomy of an LLMShare Attack
According to Push Security experts, the attack—dubbed "LLMShare"—masterfully exploits code rendering capabilities.
In ChatGPT, hackers create pages mimicking official crash notifications or system errors. Users seeking help are prompted to "fix the issue" by downloading an infected desktop application. The situation with Claude is no better: attackers pose as Apple technical support, providing step-by-step instructions where destructive terminal commands are hidden among otherwise helpful advice.
This is a classic example of how a platform's high functionality becomes its very Achilles' heel.
The End of the Whitelist Era
Similar campaigns have already been flagged by specialists at BleepingComputer and Kaspersky Lab. The problem is that the threat originates from platforms where employees are officially permitted to work. When a trusted URL can render a fake login window or trick a user into executing a dangerous command, the very concept of domain "whitelisting" loses its meaning.
It is time for businesses to face the facts: the old rules of the game are broken. Modern security requires rigorous filtering of all incoming content from neural networks, even when packaged in a polished link from OpenAI. Brand trust must not replace technical control; otherwise, your next corporate AI chat could be the last thing you do before a full system wipe.