deps(security): bump langchain-core floor to >=1.3.3 (GHSA-pjwx-r37v-7724, CVSS 8.2 HIGH)

[r266-tech]

Summary

Bump langchain-core floor from >=1.0.0 to >=1.3.3 in pyproject.toml (benchmark extra) to close GHSA-pjwx-r37v-7724 (CVSS 8.2 HIGH, published 2026-05-08).

Vulnerability

• Advisory: GHSA-pjwx-r37v-7724 — Unsafe deserialization in langchain_core.load.load() with overly broad allowed_objects. Older runtime code paths (RunnableWithMessageHistory, astream_log(), astream_events(version="v1")) call load(..., allowed_objects="all"), which lets attacker-supplied serialized constructor dicts instantiate arbitrary classes with untrusted constructor args.
• Affected: langchain-core >= 1.0.0, <= 1.3.2 (current floor >=1.0.0 permits the full vulnerable range to resolve).
• First patched: 1.3.3 (advisory tightens allowed_objects to a curated allowlist on the affected runtime surfaces).
• OpenViking exposure: langchain-core is declared under the benchmark extra alongside langchain and langchain-openai. Anyone installing the extra to run benchmarks can pull a vulnerable version until the floor is lifted.

Diff

-    "langchain-core>=1.0.0",
+    "langchain-core>=1.3.3",

(langchain>=1.0.0 and langchain-openai>=1.0.0 are unchanged — the advisory targets langchain-core only.)

Why now

Same one-line floor-bump form as recently merged #1877 (python-multipart GHSA-pp6c-gr5w-3c5g) and #1882 (litellm 5 GHSAs). Dependabot weekly cadence catches this but not for days; advisory is 2 days old at the time of this PR.

Verification

• pyproject.toml parses unchanged elsewhere; only L175 modified.
• langchain and langchain-openai floors are intentionally left at >=1.0.0 per advisory scope.

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🎫 Ticket compliance analysis ❌ 1882 - Not compliant Non-compliant requirements: Raise litellm floor to >=1.83.7 1877 - Not compliant Non-compliant requirements: Bump python-multipart floor to >=0.0.27

⏱️ Estimated effort to review: 1 🔵⚪⚪⚪⚪

🏅 Score: 95

🧪 No relevant tests

🔒 No security concerns identified

✅ No TODO sections

🔀 No multiple PR themes

⚡ No major issues detected

[github-actions]

PR Code Suggestions ✨

No code suggestions found for the PR.