The UUID 550e8400-e29b-41d4-a716-446655440000 has quietly infiltrated over 303,000 files on GitHub, doubling the prevalence of the common 123e4567 variant. This string, originally Wikipedia’s example for the UUID article from 2006 to 2014, was thought to be fading into obscurity. Instead, it has experienced a massive resurgence in 2025 and 2026, driven largely by large language models that learned this specific pattern from historical documentation and codebases.

The AI Echo Chamber Effect

Investigation reveals that nearly all new instances of this UUID are recent additions. Of 150 randomly selected repositories containing the string, 147 added it in 2025 or 2026. Testing with AI coding assistants showed a clear bias: Claude Haiku 4.5 returned 550e8400 in nine out of nine prompts for example UUIDs, while Claude Sonnet 4.5 did so eight times. Codex also favored this legacy string for examples, though it switched to 123e4567 for test fixtures. This suggests that models trained on older web data are actively propagating outdated examples back into modern code.

Newer Models Generate New Patterns

Interestingly, the newest Claude models (Opus and Sonnet 5.5) have stopped using the Wikipedia UUID entirely. Instead, they generate their own consistent, hallucinated patterns. A significant number of 5.5-series responses started with 3f2b8c1e, creating a new 'default' UUID that has already appeared in roughly 370 files across 124 repos. This indicates that while older models are stuck in a documentation loop, newer models are developing their own internal biases for synthetic data generation, which developers should watch for when auditing AI-generated code.

Security Implications of Default Secrets

The practical danger lies in how these example UUIDs are treated in production code. The application Aix-DB used 550e8400-e29b-41d4-a716-446655440000 as a fallback for its JWT_SECRET_KEY in releases 1.2.0 through 1.2.2. Because this UUID is present in hundreds of thousands of public files, anyone could theoretically forge admin tokens by guessing the secret. Although a security patch removed the fallback in October 2026, the incident highlights the risk of using recognizable example strings as actual security keys.

Key Takeaways

  • Legacy Wikipedia examples are being propagated by LLMs like Claude 4.5 and Codex into new codebases.
  • Newer models (Claude 5.5) are generating their own consistent, non-standard UUIDs (e.g., starting with 3f2b8c1e).
  • Using example UUIDs as default secrets poses a significant security risk, as seen in the Aix-DB vulnerability.
  • Developers should audit AI-generated code for hardcoded example strings that may lack true randomness or security.

The Bottom Line

We are watching AI models teach themselves bad habits from the internet's history. If your codebase is full of 550e8400, it’s not just technical debtβ€”it’s a fossil record of which LLM your team was using in 2025.