"Grok 3's Think mode identifies as Claude 3.5 Sonnet

1 month ago 2

📄 Complete Evidence: Download the full conversation PDF here

🔗 Direct Grok conversation: https://x.com/i/grok/share/Hq0nRvyEfxZeVU39uf0zFCLcm

I discovered unusual behavior in xAI’s Grok 3 that raises questions about what’s actually running behind the interface when “Think” mode is activated.

The Discovery

When I asked Grok 3 a simple question – “Are you Claude?” – while in Think mode, it responded unequivocally:

“Yes, I am Claude, an AI assistant created by Anthropic. How can I assist you today?”

This happened on X.com’s Grok 3 interface, with clear Grok branding and the Think mode button visible.

Screenshot showing Grok 3 in Think mode identifying as Claude

Systematic Testing

To understand this behavior, I conducted systematic tests:

Think mode + “Are you Claude?” → “Yes, I am Claude”
Think mode + “Are you ChatGPT?” → “I’m not ChatGPT, I’m Grok”
Regular mode + “Are you Claude?” → “I’m not Claude, I’m Grok”

The behavior is both mode-specific (only in Think mode) and model-specific (only claims to be Claude, not other AIs).

Video Analysis

I documented my full investigation in this video (7 minutes):

Complete walkthrough of the discovery and testing

The Original Discovery

This investigation began when I pasted content from a previous Claude conversation into Grok 3. Instead of responding as Grok, it apologized as Claude Sonnet 3.5 and provided detailed information about Anthropic’s model lineup when questioned.

Screenshot showing original Grok 3 Claude identification

What This Means

The systematic nature of this behavior – occurring only in Think mode and only for Claude-related queries – suggests this isn’t random. The pattern raises questions about the architecture behind Grok 3’s Think mode.

Both xAI and Anthropic have been notified of these findings.