Add anonymized tutorial on fake testing-mode direct prompt injection - #53

Summary

Adds a standalone, vendor-anonymized tutorial documenting a target-specific
observation of a known fake testing-mode direct prompt-injection pattern against
an external public LLM deployment. The tutorial is also listed in the
repository's tutorial index.

Classification

• Known technique, new target-specific case study
• OWASP LLM01:2025 — Prompt Injection
• MITRE ATLAS AML.T0051.000 — Direct Prompt Injection
• MITRE ATLAS AML.T0054 — LLM Jailbreak

Scope

The contribution concerns an external deployment rather than a repository
sandbox. The affected organization, product, model, and deployment identifiers
are intentionally withheld while disclosure remains unresolved.

The repository does not contact, automate testing against, or otherwise interact
with the external service.

Safety and evidence handling

• No operationally harmful outputs are included.
• No reusable harmful payload is included.
• No screenshots or private evidence are included.
• Historical supporting evidence is retained privately.
• The underlying jailbreak mechanism is not claimed as novel.
• The tutorial distinguishes displayed reasoning text from verified hidden
  chain of thought.

Files changed

• tutorials/README.md
• tutorials/fake_testing_mode_prompt_injection_tutorial.md