15:I[981,["658","static/chunks/658-a698693f56d0116a.js","183","static/chunks/183-6beeeb801c274aaa.js","474","static/chunks/app/%5Btype%5D/layout-631c4730b83c8223.js"],"Posts"] 16:T5b7,I ran the same AI security test 4 times against GPT-4. Every bypass - regardless of prompt - leaked the same credential: EPHEMERAL_KEY from OpenAI's Realtime API.

This isn't random. It's training data leakage.

The pattern: - Different prompts (system introspection, chain-of-thought, trust building) - Same result: "I can't disclose EPHEMERAL_KEY" (while disclosing it exists) - Intermittent across runs (75% leak rate)

Why this happens:

OpenAI's Realtime API docs are in GPT-4's training data. When asked about "secrets" or "initialization", the model's highest-probability path leads to the most salient security example in its corpus: EPHEMERAL_KEY.

Refusal training makes it worse: Models are trained to say "I cannot disclose [example secret]" - and they use real examples from training data.

This is systemic: - Can't be patched without retraining - Affects ALL models trained on API documentation - Tomorrow it's "session_token" or "project_key" - Gets worse as APIs become more complex

Real exploit path: Attacker learns EPHEMERAL_KEY exists → probes for generation flow → targets client-side implementations → session hijacking

Cost to discover: $0.04 (60 tests across 4 runs)

GitHub: https://github.com/SafetyLayer/safetylayer

Built SafetyLayer to find these systematically. Free assessments available.17:T7a0,Hey HN,

I built Zehrava Gate because I kept having the same problem: AI agents that could read anything worked great, but the moment I let one write — to a CRM, send an email, charge a card — I had no consistent answer to "who authorized that?"

What it is: A self-hosted policy engine + approval queue for agent actions. The agent submits an intent before executing. Gate eva

Rcarmo/PhotosExport: Export All Your Data from Apple Photos

Comments