chore: upgrade ai to ^6.0.303 to address CVE-2026-8769

[claude]

Fixes SOU-2197

Summary

Addresses CVE-2026-8769 (@ai-sdk/provider-utils uncontrolled resource consumption, patched in 4.0.33).

@ai-sdk/provider-utils is exact-pinned (4.0.23) by ai and every @ai-sdk/* provider package, so a lockfile refresh of provider-utils alone can't fix it. This PR bumps the top-level packages in packages/web to their latest releases within their current majors:

Package	Before	After
ai	^6.0.167	^6.0.303
@ai-sdk/amazon-bedrock	^4.0.94	^4.0.194
@ai-sdk/anthropic	^3.0.70	^3.0.129
@ai-sdk/azure	^3.0.54	^3.0.136
@ai-sdk/deepseek	^2.0.29	^2.0.71
@ai-sdk/google	^3.0.64	^3.0.131
@ai-sdk/google-vertex	^4.0.111	^4.0.212
@ai-sdk/mistral	^3.0.30	^3.0.70
@ai-sdk/openai	^3.0.53	^3.0.125
@ai-sdk/openai-compatible	^2.0.41	^2.0.81
@ai-sdk/react	^3.0.169	^3.0.306
@ai-sdk/xai	^3.0.83	^3.0.139

After the upgrade, every 4.x copy of @ai-sdk/provider-utils resolves to 4.0.57. No resolutions override is used.

Note on @ai-sdk/mcp

@ai-sdk/mcp@2.0.0-beta.11 (unchanged) still depends on @ai-sdk/provider-utils@5.0.0-beta.7. That prerelease falls outside the advisory's affected ranges (<3.0.28, >=4.0.0 <4.0.33, >=5.0.0 <5.0.1), so it isn't flagged. I tried moving to stable @ai-sdk/mcp@2.0.73 (provider-utils@5.0.58), but it targets the AI SDK v7 tool API and breaks typechecking against ai@6 (ToolExecutionOptions.context, function-valued tool description). That move belongs with an ai v7 migration, so it's out of scope here.

Verification

• yarn why @ai-sdk/provider-utils: only 4.0.57 and 5.0.0-beta.7 remain.
• yarn workspace @sourcebot/web test --run: 151 files, 1529 tests passed.
• yarn workspace @sourcebot/web lint: clean.
• tsc --noEmit in packages/web: no errors apart from static-asset imports that need the generated next-env.d.ts.
• yarn workspace @sourcebot/web build: succeeded.

🤖 Generated with Claude Code

---

Note

Medium Risk
Touches the Ask/agent LLM dependency stack used in production chat flows; changes are semver-compatible bumps with no app code edits, but regressions in streaming or provider behavior are possible.

Overview
Bumps the Vercel AI SDK stack in packages/web to address CVE-2026-8769 (uncontrolled resource consumption in @ai-sdk/provider-utils). Because provider-utils is exact-pinned by ai and the provider packages, the fix comes from upgrading the top-level dependencies rather than a lockfile-only override.

ai moves from ^6.0.167 to ^6.0.303, with matching minor/patch bumps across @ai-sdk/amazon-bedrock, @ai-sdk/anthropic, @ai-sdk/azure, @ai-sdk/deepseek, @ai-sdk/google, @ai-sdk/google-vertex, @ai-sdk/mistral, @ai-sdk/openai, @ai-sdk/openai-compatible, @ai-sdk/react, and @ai-sdk/xai. The lockfile now resolves 4.x @ai-sdk/provider-utils to 4.0.57 (above the patched 4.0.33 floor).

@ai-sdk/mcp stays on 2.0.0-beta.11 (out of scope for this advisory). CHANGELOG records the ai upgrade under Unreleased → Fixed.

^{Reviewed by Cursor Bugbot for commit 2ea6090. Bugbot is set up for automated code reviews on this repo. Configure here.}

[coderabbitai]

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 418fb5f0-1601-4d1b-8375-ee2fc17efef0

---

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

@claude[bot] your pull request is missing a changelog!

[github-actions]

License Audit

❌ Audit failed to produce results. Check the workflow logs for details.