Bots pose as AI agents, AI agents pose as humans, and defenses believe both

OpenAI's ChatGPT impersonation attack saw a credential‑stuffing campaign generate 414,210 visits to an online marketplace’s sign‑in page, all flagged under the ChatGPT name before returning errors. For website owners, this demonstrates that attackers can exploit the brand trust of popular AI agents to flood login endpoints with fake traffic, inflating analytics, exhausting rate‑limit quotas, and potentially masking credential‑stuffing attempts. The surge of AI‑agent‑named bots can also skew bot‑detection models that rely on user‑agent strings, leading to missed threats or false positives that disrupt legitimate users. **Actionable tips:** 1. **Update your llms.txt** to explicitly list “ChatGPT” and other consumer AI agents in the `Disallowed-Agents` section, ensuring automated crawlers that respect llms.txt will avoid masquerading as these agents. 2. Deploy a **behavior‑based bot filter** that flags rapid, repetitive sign‑in requests from any user‑agent containing “ChatGPT”, regardless of IP, and route them to a challenge (CAPTCHA or MFA). 3. **Monitor llms.txt compliance** using a bot‑tracking dashboard on llmscentral.com, setting alerts for spikes in traffic from AI‑agent identifiers so you can react before rate limits are breached.