2629aabcc5
4 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
Aaron D. Lee
|
2629aabcc5 |
Fix 12 security findings from adversarial audit
CRITICAL: - #1+#2: Consistency proof verification no longer a stub — implements actual hash chain reconstruction from proof hashes, rejects proofs that don't reconstruct to the expected root. GossipNode._verify_consistency now calls verify_consistency_proof() instead of just checking sizes. - #3: Remove passphrase.lower() from KDF — was silently discarding case entropy from mixed-case passphrases. Passphrases are now case-sensitive as users would expect. - #4: Federation gossip now applies record_filter (trust store check) on every received record before appending to the log. Untrusted attestor fingerprints are rejected with a warning. - #5: Killswitch disables all logging BEFORE activation to prevent audit log from recording killswitch activity that could survive an interrupted purge. Audit log destruction moved to position 4 (right after keys + flask secret, before other data). HIGH: - #6: CSRF exemption narrowed from entire dropbox blueprint to only the upload view function. Admin routes retain CSRF protection. - #7: /health endpoint returns only {"status":"ok"} to anonymous callers. Full operational report requires authentication. - #8: Metadata stripping now reconstructs image from pixel data only (Image.new + putdata), stripping XMP, IPTC, and ICC profiles — not just EXIF. - #9: Same as #6 (CSRF scope fix). MEDIUM: - #11: Receipt HMAC key changed from public upload token to server-side secret key, making valid receipts unforgeable by the source or anyone who captured the upload URL. - #12: Docker CMD no longer defaults to --no-https. HTTPS with self-signed cert is the default; --no-https requires explicit opt-in. - #14: shred return code now checked — non-zero exit falls through to the zero-overwrite fallback instead of silently succeeding. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> |
||
|
Aaron D. Lee
|
f557cac45a |
Implement 6 evidence lifecycle features
1. Client-side SHA-256 in drop box: browser computes and displays file fingerprints via SubtleCrypto before upload. Receipt codes are HMAC-derived from file hash so source can verify correspondence. Source sees hash before submitting. 2. Drop box token persistence: replaced in-memory dict with SQLite (dropbox.db). Tokens and receipts survive server restarts. Receipt verification now returns filename, SHA-256, and timestamp. 3. RFC 3161 trusted timestamps + manual anchors: new federation/anchors.py with get_chain_head_anchor(), submit_rfc3161(), save_anchor(), and manual export format. CLI: `soosef chain anchor [--tsa URL]`. A single anchor implicitly timestamps every preceding chain record. 4. Derived work lineage: attestation metadata supports derived_from (parent record ID) and derivation_type (crop, redact, brightness, etc.) for tracking edits through the chain of custody. 5. Self-contained evidence package: new soosef.evidence module with export_evidence_package() producing a ZIP with images, attestation records, chain data, public key, standalone verify.py script, and README. 6. Cold archive export: new soosef.archive module with export_cold_archive() bundling chain.bin, verisoo log, LMDB index, keys, anchors, trusted keys, ALGORITHMS.txt documenting all crypto, and verification instructions. Designed for OAIS (ISO 14721) alignment. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> |
||
|
Aaron D. Lee
|
171e51643c |
Add extract-then-strip EXIF pipeline for attestation intake
Resolves the tension between steganography (strip everything to protect sources) and attestation (preserve evidence of provenance): - New soosef.metadata module with extract_and_classify() and extract_strip_pipeline() — classifies EXIF fields as evidentiary (GPS, timestamp — valuable for proving provenance) vs dangerous (device serial, firmware — could identify the source) - Drop box now uses extract-then-strip: attests ORIGINAL bytes (hash matches what source submitted), extracts evidentiary EXIF into attestation metadata, strips dangerous fields, stores clean copy - Attest route gains strip_device option: when enabled, includes GPS/timestamp in attestation but excludes device serial/firmware - Stego encode unchanged: still strips all metadata from carriers (correct for steganography threat model) The key insight: for stego, the carrier is a vessel (strip everything). For attestation, EXIF is the evidence (extract, classify, preserve selectively). Both hashes (original + stripped) are recorded so the relationship between raw submission and stored copy is provable. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> |
||
|
Aaron D. Lee
|
9431033c72 |
Implement 7 real-world scenario features (Round 4)
1. Source drop box: token-gated anonymous upload with auto-attestation, EXIF stripping, receipt codes, and self-destructing URLs. New /dropbox blueprint with admin panel for token management. CSRF exempted for source-facing upload routes. 2. Investigation namespaces: attestation records tagged with investigation label via metadata. Log view filters by investigation with dropdown. Supports long-running multi-story workflows. 3. Scale fixes: replaced O(n) full-scan perceptual hash search with LMDB find_similar_images() index lookup. Added incremental chain verification (verify_incremental) with last_verified_index checkpoint in ChainState. 4. Deep forensic purge: killswitch now scrubs __pycache__, pip dist-info, pip cache, and shell history entries containing 'soosef'. Runs before package uninstall for maximum trace removal. 5. Cross-org federation: new federation/exchange.py with export_attestation_bundle() and import_attestation_bundle(). Bundles are self-authenticating JSON with investigation filter. Import validates against trust store fingerprints. 6. Wrong-key diagnostics: enhanced decrypt error messages include current channel key fingerprint hint. New carrier_tracker.py tracks carrier SHA-256 hashes and warns on reuse (statistical analysis risk). 7. Selective disclosure: ChainStore.selective_disclosure() produces proof bundles with full selected records + hash-only redacted records + complete hash chain for linkage verification. New `soosef chain disclose -i 0,5,10 -o proof.json` CLI command for court-ordered evidence production. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> |