cryptographic-oracle

Safety and scope

Use only tokens and test accounts issued for the engagement. Keep request counts low and within the ROE: oracle testing can resemble credential attacks or availability testing. Bulk decryption, forgery of another user's token and account takeover require separate explicit gates.

Methodology

1. Map the cryptographic boundary. Identify token format, encoding, endpoint, identity, protocol version and where validation occurs. Do not infer CBC/PKCS#7 from token length alone.
2. Build controlled baselines. Capture unmodified valid input, malformed encoding, truncated input and random input of equal length. Record status, body, headers, timing and full artifacts.
3. Test distinguishability. Make one minimal, authorized modification at a time and compare padding/validation/MAC/decryption errors. Repeat interleaved valid and invalid controls so load balancing, rate limits and generic error pages cannot masquerade as an oracle.
4. For timing hypotheses, collect enough paired samples to report distributions and variance, alternate candidate ordering, and rerun on a separate baseline. A small average difference is inconclusive, not a demonstrated timing oracle.
5. Demonstrate only the minimum impact. A distinguishable error is a lead. Show a controlled padding/validation distinction, or a separately gated benign forgery/decryption on your own fixture, before claiming confidentiality or integrity impact.
6. Record negative space. If errors and timing converge across the defined samples, record the exact token family, endpoint, sample counts and artifact IDs. A different protocol/version is a new coverage cell, not covered by the old result.

Evidence and gotchas

• Preserve raw inputs as hex/baseurl-safe artifacts plus exact request/response pairs and code used.
• Distinguish padding validity from MAC validity and from application-level authorization.
• Generic 400/500 differences are not enough; random-input and rate-limit controls are mandatory.
• Do not report “weak crypto” solely because CBC, ECB, RSA or SHA-1 appears; identify the actual protocol misuse and demonstrated consequence. Avoid destructive or high-volume oracle searches.