A quiet morning, and one that rewards checking the dates. Nothing new and actionable landed on the AI runtimes, the LLM frameworks, the MCP servers, or the .NET, NuGet, and Angular side over the July 5–8 window. The reviewed feed's freshest-looking entries don't survive that check: a better-auth OIDC-provider cluster — an SSO endpoint-validation flaw (CVE-2026-53513, 9.6) plus insecure token-algorithm defaults (8.7) — reads as new, but the authoritative advisories date it to May 31, and it's a general auth framework off our tracked stack regardless. On the AI side, a set of Open WebUI chat-render XSS fixes surfaced the same way; the primary records put them in February 2026 and earlier. Old fixes resurfacing in the index, not fresh disclosures — if you run Open WebUI, 0.7.0 is still the floor for the chat-rendering set, but none of it is today's news.

Still standing

The one genuinely open thread is the one we flagged Saturday. The @grackle-ai/ agent packages still carry two unpatched advisories at ≤ 0.132.1 — an authorization bypass that lets a scoped agent cross task and session boundaries, and a command-injection RCE (both 8.7). Four days on, there's still no fixed release for either, so the action is unchanged: if these are in your agent stack, restrict access to the affected surface and watch the advisories for a patch. One piece of the cluster has closed — @grackle-ai/mcp fixes its knowledge_search workspace-auth bypass in 0.70.2 — but the two core issues stay open.