fix(deps): resolve dependabot security alerts in integration lockfiles

[Abhijeet Prasad (AbhiPrasad)]

Summary

Clears six open Dependabot security advisories by bumping anyio, cryptography and idna in the two deprecated integration compatibility packages. Both lockfiles had drifted behind py/uv.lock, which already carried patched versions of all three.

Package	integrations/adk-py	integrations/langchain-py
anyio	4.13.0 → 4.14.2	4.13.0 → 4.14.2
cryptography	48.0.1 → 50.0.1	49.0.0 → 50.0.1
idna	3.13 → 3.20	3.13 → 3.20
pyopenssl	26.2.0 → 26.4.0	—

pyopenssl moves only because cryptography 50 requires it.

Advisories cleared

Advisory	Severity	Package	Issue
GHSA-82r6-8w77-94w6	Critical	anyio	TLSStream IDNA 2003 host name encoding enables TLS certificate spoofing
GHSA-g6cj-pr64-35w5	High	cryptography	PKCS#7 EnvelopedData decryption exposes a Bleichenbacher oracle
GHSA-jwv3-5hgf-82ww	High	cryptography	Duplicate self-signed intermediates cause exponential path-building
GHSA-5p39-cfhj-2xmp	Moderate	anyio	Process-pool workers block indefinitely on undrained stderr
GHSA-m2h6-j472-rp4c	Moderate	cryptography	Verifier accepts wildcard DNS names, escaping permittedSubtrees
GHSA-65pc-fj4g-8rjx	Moderate	idna	Crafted input to idna.encode() bypasses the CVE-2024-3651 fix

Approach

Used targeted uv lock --upgrade-package anyio --upgrade-package cryptography --upgrade-package idna in each package directory rather than a full re-lock, so the diff is limited to the four packages above.

Validation

Re-audited both lockfiles against OSV after the bumps — both scan clean, 0 known vulnerabilities.

Ran the exact CI step list from adk-py-test.yaml and langchain-py-test.yaml for each package (uv sync, ruff check, pytest, import smoke test). All green:

• integrations/adk-py — ruff clean, src/tests/test_reexports.py 2 passed, braintrust_adk / setup_braintrust import OK
• integrations/langchain-py — ruff clean, src/tests 2 passed, braintrust_langchain / BraintrustCallbackHandler import OK

Not fixed — no patched release exists upstream

Two advisories in py/uv.lock have no fixed version available, so there is nothing to bump to:

• diskcache 5.6.3 — GHSA-w8v5-vhqr-4h9v (Moderate), unsafe pickle deserialization. 5.6.3 is the latest release on PyPI, from 2023.
• nltk 3.10.3 — GHSA-8mgp-746c-j5xp (High), model-artifact APIs bypass path security. 3.10.3 is the latest release and is marked last_affected.

Both are transitive dev-only dependencies of the lint dependency group (via dspy and llama-index-core respectively). Neither is a runtime dependency of the published braintrust package.

Options, if you want them off the alert list: dismiss them in the Dependabot UI as "no fix available", or drop dspy / llama-index-core from the shared lint group the way crewai already is (see the existing precedent comment in py/pyproject.toml). Left alone here to avoid changing lint coverage unprompted.

Note on how these were found

The gh token available in this environment lacks vulnerability_alerts=read, so the alert list could not be read directly — the REST endpoint returns 403, and the GraphQL vulnerabilityAlerts fallback silently returns totalCount: 0 rather than erroring. The findings above were reproduced by parsing all three tracked lockfiles and querying the OSV API, which mirrors the same GHSA identifiers Dependabot uses.

The push that created this branch surfaced GitHub's own count for the default branch: 4 vulnerabilities (2 critical, 2 moderate). That matches anyio 4.13.0 exactly — it carries one critical (GHSA-82r6-8w77-94w6) and one moderate (GHSA-5p39-cfhj-2xmp) advisory, and it is pinned at 4.13.0 in both integration lockfiles: 2 x critical + 2 x moderate = 4. On that reading, this PR clears all four open alerts.

The cryptography and idna bumps are then additional hardening that OSV flags but Dependabot is not currently alerting on. They cost nothing here, since both packages had already advanced past these advisories in py/uv.lock.

Since the alert list could not be read directly, please sanity-check the above against the real Dependabot page when reviewing.

🤖 Generated with Claude Code

Co-authored by StarfolkAI (@starfolkai)[bot]

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-24T20:57:15.517784Z	3311bd3	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "Codex (@codex) review" or "Codex (@codex) security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.