A critical CMS bug, a new switch for MCP servers, and three platform changes from Friday and the weekend.
- GitHub says TinaCMS versions up to 3.13.0 let a crafted link steer the admin preview to an attacker’s page.
- GitHub Copilot for JetBrains can now stop MCP servers from starting automatically, and drops support for IDE 2025.1.
- AWS says Amazon Bedrock now returns reasoning summaries for OpenAI models.
- GitHub says CodeQL 2.27.2 changes how Go code is modelled, which can affect custom queries.
- AWS says Security Hub can now export findings to S3 as CSV or OCSF JSON.
Security
TinaCMS admin preview can be pointed at an attacker’s page
GitHub rates this advisory (CVE-2026-108261) critical. It says the TinaCMS admin accepts a preview address from the URL fragment, so a crafted link opens an attacker’s page inside the admin, and that page can then send queries and changes using the signed-in editor’s access. The advisory says the payload sits in the fragment, so it never reaches the server or its logs. It lists tinacms up to 3.13.0 (fixed in 3.14.0) and @tinacms/app up to 2.5.13 (fixed in 2.5.14). It hits any site that serves the TinaCMS admin.
This week. Upgrade tinacms to 3.14.0 and @tinacms/app to 2.5.14, then rebuild and redeploy the admin. On self-hosted setups, review recent content changes and reset editor passwords, since the advisory says the attacker can read the authentication collection.
AI
Copilot for JetBrains can stop MCP servers from starting on their own
GitHub’s 10 October changelog adds a setting that turns off automatic MCP server startup for Copilot and Claude, so configured tools only become active when you choose. The same release ends support for JetBrains IDE 2025.1, so the plugin now needs 2025.2 or later. It hits teams that share MCP configuration across a repository, and anyone still on an older IDE.
This week. Check that every developer is on JetBrains IDE 2025.2 or later before the plugin update lands. In repositories that carry MCP server definitions, turn the automatic startup off so a server runs only after someone has looked at it.
Bedrock now returns reasoning summaries for OpenAI models
AWS says Amazon Bedrock now supports the reasoning.summary parameter for OpenAI models through the Responses API. The summary comes back in the summary array of the reasoning output item, next to the answer. AWS says it works in-Region and with geographic and global cross-Region inference. It hits teams that call OpenAI models on Bedrock and want something to show in logs or to users.
This week. On one Bedrock workload that calls an OpenAI model, send reasoning.summary and read the summary array. Decide before you store it whether the text may contain customer data, and apply the same retention rules as your other model logs.
Platforms
CodeQL 2.27.2 changes how Go code is modelled
GitHub says CodeQL 2.27.2 moves the Go control-flow graph onto the shared CFG library, adding nodes for assignments, parameters and results, range statements, and deferred calls, and leaving out nodes that cannot be reached. GitHub says this changes CFG nodes and edges. The release also adds a C++ regular-expression parser and analysis changes for Rust and JavaScript/TypeScript. It hits teams that run their own CodeQL queries against Go.
This week. Run your custom Go query packs against CLI 2.27.2 on a test branch and compare the results with your current version before you update CI.
Security Hub can export findings to S3
AWS says you can now start an export from any findings page in the Security Hub console and have the results delivered to an S3 bucket in your account, as CSV or as JSON in the Open Cybersecurity Schema Framework (OCSF) format. AWS says this is available in every Region where Security Hub is. It hits security and compliance teams that now pull findings out by hand.
This week. Run one export in each format. If someone on your team copies findings into spreadsheets for audits, check whether the CSV export covers that job.
Leave a Reply