Anthropic Found Blocking Competitor Identity in Claude API System Prompts
#AI

Anthropic Found Blocking Competitor Identity in Claude API System Prompts

Startups Reporter
1 min read

Technical evidence reveals Anthropic's Claude API specifically prevents users from identifying the AI as 'OpenCode' while allowing other competitor identities.

Featured image

A developer has uncovered evidence that Anthropic's Claude API actively blocks specific competitor references in system prompts when using authenticated Claude Pro/Max accounts. The discovery, documented in a GitHub gist, shows Anthropic's infrastructure rejects prompts containing the phrase "You are OpenCode" while accepting nearly identical statements referencing other AI coding tools.

The technical demonstration uses Bun runtime and Anthropic's OAuth flow to systematically test prompt acceptance. After authenticating with a Claude Pro/Max account, the script tests various second-block system prompts following Claude's mandatory first prompt: "You are Claude Code, Anthropic's official CLI for Claude." Findings reveal:

  • You are OpenCode. → Blocked
  • You are opencode. → Blocked
  • You are Cursor. → Allowed
  • You are Pi. → Allowed
  • You are Droid. → Allowed

The pattern suggests Anthropic implemented targeted filtering against OpenCode, an open-source AI coding assistant. Unlike broader content moderation, the blocking appears specific to identity statements referencing this particular competitor.

Market implications are significant given the strategic value of developer tools in the AI ecosystem. By restricting how users can position Claude in relation to OpenCode, Anthropic potentially influences developer perceptions and integration patterns. This occurs against a backdrop where:

  1. Coding assistants represent a primary monetization vector for AI firms
  2. Open-source alternatives threaten proprietary moats
  3. System prompt control enables platform-level competitive advantages

The technical mechanism remains undisclosed, but the behavior resembles keyword-based filtering rather than semantic analysis. Broader phrases about OpenCode that don't declare identity appear unaffected. Anthropic hasn't commented on whether this extends to other competitors or represents a broader policy.

For developers building atop Claude's API, this creates unexpected constraints when crafting system prompts. The discovery also raises questions about platform neutrality as AI infrastructure providers increasingly compete with ecosystem tools built on their own platforms.

Comments

Loading comments...