diff --git a/CHANGELOG.md b/CHANGELOG.md
index 3fa8827..781d252 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,40 @@
 # Changelog
 
+## [Unreleased] - 2026-07-15
+
+### Security
+
+Multi-round security-hardening pass (audit → adversarial review → fix,
+followed by standing red-team/blue-team sparring rounds) ahead of SOC 2
+Type II control-mapping work. See [SECURITY.md](SECURITY.md#security-hardening--soc-2-type-ii-readiness)
+for the full readiness summary. Highlights:
+
+- **`verify()` on TSA timestamps was not actually verifying the TSA's
+  attestation** — it compared `sha256(data)` against a caller-supplied
+  `message_imprint` field instead of the hash the TSA itself signed. Now
+  DER/CMS-decodes the real response and checks the TSA-attested hash and
+  `SignerInfo` signature.
+- **SSRF**: outbound TSA requests now require `https://` and resolved hosts
+  are checked against private/loopback/link-local/metadata-endpoint ranges.
+- **Timing side-channels** in the pure-Python secp256k1 signer: modular
+  inverse moved to fixed-shape exponentiation; scalar multiplication moved
+  to a constant-iteration schedule.
+- **Key hygiene**: `destroy()` now overwrites the private-key buffer in
+  place instead of just dropping the reference.
+- **Execution/settlement integrity**: execution attestations are now
+  cross-validated against the commitment's authorised trade terms
+  (qty/price/symbol/side); broker settlement acknowledgements are now
+  cryptographically authenticated instead of accepted as a free-form string.
+- **Identity binding**: new `AccountKeyRegistry` (`identity.py`) closes a
+  gap where any self-signed key could pass every check with no proof it
+  belonged to the account holder.
+- **Concurrency**: audit-log JSONL/SQLite-index writes are now lock-guarded
+  against a race that could silently overwrite a duplicate order/phase
+  record.
+- DoS bounds on unbounded JSON/HTTP-response reads; silent exception
+  swallowing removed from the audit-rebuild and TSA request paths.
+- Every fix carries a dedicated regression test; suite grew to 137 tests.
+
 ## [0.1.0] - 2026-06-03
 
 ### Added
diff --git a/README.md b/README.md
index 2489684..9ac76eb 100644
--- a/README.md
+++ b/README.md
@@ -200,6 +200,8 @@ Each phase lands in the audit log under `DECISION_COMMITMENT`, `EXECUTION_ATTEST
 
 See [SECURITY.md](SECURITY.md) for the full threat model, assumptions, and how to report a vulnerability. The full technical white paper — the CVE-2025-59536 problem, the design, and honest scope — is in [`docs/WHITEPAPER.md`](docs/WHITEPAPER.md).
 
+**Security hardening.** Protocol-C went through a multi-round audit → adversarial-review → fix cycle plus standing red-team/blue-team sparring in July 2026 — TSA verification, SSRF, signer timing side-channels, key hygiene, execution/settlement integrity, identity binding, and audit-log concurrency. Not SOC 2 certified — see [SECURITY.md#security-hardening--soc-2-type-ii-readiness](SECURITY.md#security-hardening--soc-2-type-ii-readiness) for exactly what's covered and what isn't.
+
 ## Honest about "quantum-safe"
 
 Read this part. "Quantum-safe" here is a **temporal** argument, **not** post-quantum cryptography:
diff --git a/SECURITY.md b/SECURITY.md
index c039748..9dfbf1a 100644
--- a/SECURITY.md
+++ b/SECURITY.md
@@ -69,6 +69,42 @@ This public package is **CSPRNG-only**: all entropy comes from
 `secrets.token_bytes`. The quantum-hardware entropy variant is a separate,
 private project and ships no code here.
 
+## Security hardening & SOC 2 Type II readiness
+
+Protocol-C is not SOC 2 certified — no independent auditor has issued a
+report. What's here is an honest account of the engineering work done so
+far toward that posture, so anyone evaluating the library for a
+compliance-relevant use case can see exactly what's been checked and
+what's still open.
+
+**2026-07-15 hardening pass** (tracked in [PR #8](https://github.com/DBarr3/protocol-c/pull/8)):
+a multi-round audit → adversarial review → fix cycle, followed by standing
+red-team/blue-team sparring rounds against the hardened surface. Every
+fix carries a dedicated regression test (suite: 137 tests). Summary —
+full detail in [CHANGELOG.md](CHANGELOG.md#unreleased---2026-07-15):
+
+| Area | What was found and fixed |
+|---|---|
+| TSA verification | `verify()` wasn't checking the TSA's actual signed attestation — now parses and verifies the real CMS/TSTInfo structure. |
+| Network (SSRF) | Outbound TSA requests now require HTTPS and reject private/loopback/link-local/metadata-endpoint hosts. |
+| Timing side-channels | Modular inverse and scalar multiplication in the secp256k1 signer moved to fixed-shape/constant-iteration implementations. |
+| Key hygiene | `destroy()` now zeroes the private-key buffer in place. |
+| Business-logic integrity | Execution attestations are now cross-checked against the commitment's authorised trade terms; broker settlement acknowledgements are now cryptographically authenticated. |
+| Identity binding | New `AccountKeyRegistry` closes a gap where any self-signed key could pass every check with no proof of account ownership. |
+| Concurrency | Audit-log writes are now lock-guarded against a race that could silently overwrite a duplicate record. |
+| Availability | DoS bounds added on unbounded JSON/HTTP-response reads. |
+| Observability | Silent exception swallowing removed from the audit-rebuild and TSA request paths in favor of structured logging. |
+
+This addresses SOC 2 control areas most relevant to a signing/attestation
+library — **change integrity** (tamper detection, non-repudiation),
+**logical access** (identity binding on signers), and **availability**
+(DoS bounds, graceful failure) — but a hardening pass is not a
+certification. Still open before any real SOC 2 Type II claim: an
+independent third-party audit, formal control documentation, and
+continuous-monitoring evidence over the required observation period. If
+you're evaluating Protocol-C for a compliance-relevant deployment, treat
+this section as "here's what's been checked," not "this is certified."
+
 ## Cryptographic implementation notes
 
 - **Signer:** original pure-Python secp256k1 ECDSA in `ephemeral_signer.py`,