AI Coding Tools Weekly, October 26-31, 2026: Copilot Fusion GA and Claude Code's Opus 4.8 Fast Mode Deadline

Executive Summary

• GitHub published 20 Copilot entries on October 28, tied to its Universe event. Copilot Fusion and GitHub Agentic Workflows are generally available, and Agent Merge now runs in the cloud on pull requests.
• Claude Code reached 2.1.316. Fast mode on Opus 4.8 is deprecated and will be removed on November 30, 2026, after which those requests return an error.
• Codex went from 0.164.0 to 0.166.0, adding a Vertex AI provider, GPT-6.1 Terra and Memory v2 for fresh profiles.
• Help Net Security reported research on the embedded browsers of five AI coding tools. Two are reported as fixed, and the details are not yet confirmed by vendor advisories.
• Copilot assisted permissions are GA, GPT-6.1 Terra is rolling out, and server-to-server authentication is in public preview.

GitHub Copilot and the Universe Batch

Copilot Fusion is generally available

Per the Fusion changelog entry, Fusion is a model option that routes and orchestrates models for each task so you don't pick one yourself. You choose an effort level (Standard, High or Max) and Fusion picks a workflow per prompt:

• Single: one model handles the task.
• Cascade: an efficient model drafts, and a quality check accepts the result or escalates to a stronger model.
• Critique: one model drafts, another reviews, and the original revises.

It is available on all Copilot plans in VS Code, the Copilot app, Copilot CLI and JetBrains IDEs. You are billed at each model's standard rate, with a limited-time 30% discount on model costs for paid plans. GitHub claims Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max at about half the average API cost per task. That is GitHub's own figure, so test it on your workloads. Admins can restrict which models Fusion may use and set it as the enterprise default.

GitHub Agentic Workflows reach GA

The general availability announcement covers the GitHub Next project that lets you write automations in natural-language Markdown, compiled into standard Actions YAML. Typical jobs are issue triage, CI failure analysis, documentation updates and test writing.

Agents get read-only permissions by default and run in a sandboxed container behind a firewall, and their outputs pass safe-outputs validation and a threat detection job. New since preview: self-hosted ARC runners, model routing by task difficulty, per-workflow model provider selection and micro-VM isolation. It is available on Free, Team and Enterprise, and the default Copilot CLI engine needs a Copilot license.

Agent Merge and Max code review

Agent Merge now runs in the cloud on GitHub pull requests. Pick "Watch and fix" in the merge box and Copilot works on failing workflows, merge conflicts or both, and it can pair with auto-merge. It needs a paid Copilot plan.

Copilot code review gets a Max effort level in public preview, which runs several independent review agents across providers and reconciles their findings, aimed at risky changes like auth logic. Copilot approvals are now GA, off by default, and an approval is dismissed when new commits are pushed.

Assisted permissions are GA

Per the GitHub changelog, assisted permissions is a permission mode in Copilot CLI, the Copilot app and VS Code. When Copilot wants to run a command, access a file path or open a URL, a language model reviews the request. Anything it classifies as clearly safe proceeds automatically, and everything else still asks you.

• Enable it with /permissions in the CLI, the Tool permissions picker in the app, or the permissions picker in VS Code.
• Enterprise admins can block Allow All while keeping Assisted, disable Assisted entirely, or set the default mode for new sessions. If the default is unavailable, sessions fall back to manual approval.

It is a middle path between approving every prompt and running with everything allowed, but the classifier is a model, so keep deny rules for anything destructive.

GPT-6.1 Terra

GitHub's GPT-6.1 Terra entry says the model is rolling out gradually to Copilot Pro, Pro+, Max, Business and Enterprise. It appears in the model picker across VS Code, Visual Studio, Copilot CLI, the cloud agent, github.com, GitHub Mobile, JetBrains, Xcode and Eclipse, and is billed at provider list pricing under usage-based billing. Business and Enterprise admins control access through the model policy in Copilot settings.

Server-to-server authentication (public preview)

The GitHub changelog entry describes GitHub Apps installed on an organization that can now make Copilot requests through the Copilot SDK without a signed-in user's credentials. The identity belongs to the workload instead of one person.

1. Give the GitHub App the read-only Copilot Requests organization permission.
2. Install the app on the organization.
3. Pass a short-lived installation access token to the Copilot SDK.

Usage is billed to the organization that owns the installation, and its model policies, budgets and spending controls apply. Owners can see app-attributed usage in AI usage CSV reports via the app_client_id field and revoke access by suspending or uninstalling the app. The changelog gives no general availability date.

Other Copilot changes

• Copilot SDK becomes Microsoft Copilot SDK: no immediate changes for existing users, with package names and docs migrating in phases and no dates given.
• GitHub Enterprise now includes access to Copilot. Check the entry for the plan details.
• Issue artifacts entered public preview and a security extension arrived for the Copilot app.
• Copilot Automations gained new tools, triggers and cost controls, plus VS Code management and outcome-based self-improvement.

The GitHub Copilot changelog had nothing dated after October 28 as of October 31.

Claude Code: 2.1.311 to 2.1.316

Opus 4.8 fast mode has a deadline

According to the Claude Code release notes, 2.1.316 deprecates fast mode for Opus 4.8 as of October 30, 2026, with removal on November 30, 2026. After removal, fast mode requests on Opus 4.8 return an error and do not fall back to standard Opus 4.8. The notes say to migrate to Opus 5.5 to keep the speedup. If you have fast mode pinned in scripts, shared settings or CI jobs, those calls will fail rather than slow down.

Permissions and security

• Write boundary (2.1.314): permissions.blockWritesOutsideWorkingDirectories makes Write, Edit and NotebookEdit refuse paths outside your working directories in every permission mode.
• Restricted sessions (2.1.316): --restricted sessions no longer upload files from outside the working directories when a Read allow rule covered them.
• Gateway model checks (2.1.316): the gateway now checks every model a request names against availableModels, deniedModels and models, returning a 400 that names the blocking field.
• Managed settings (2.1.316): server-managed settings for Team and Enterprise logins fetched from Anthropic's API now apply without an approval prompt, and a stored copy that does not verify is refused.

Effort and ultrathink

• ultrathink in the interactive terminal now runs its turn at xhigh effort on Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1 through Anthropic's API, and the rainbow highlight only shows where the keyword can actually raise effort (2.1.311). In VS Code, ultrathink also runs its turn at xhigh effort, and CLAUDE_CODE_DISABLE_ULTRATHINK turns the keyword off (2.1.316).
• CLAUDE_CODE_SUBAGENT_EFFORT sets the effort level subagents run at when nothing else names one, instead of inheriting the session's level (2.1.316).
• Sessions no longer fail with "Invalid effort level" when an admin caps a model's effort. Claude Code lowers the effort, retries and reports the level it used (2.1.314).

Fewer freezes and failed runs

Version 2.1.311 (October 26) is a large release, mostly fixes:

• Fixed long freezes at startup and after file reads or edits when a CLAUDE.md file has a very long list item.
• Write and Edit no longer stall for up to 5 seconds on edits that change thousands of lines.
• Subagents can now use MCP tools from servers that finish connecting after the subagent starts, and SubagentStop hooks run once instead of about twice a minute.
• Standard-context sessions that cross 200K tokens now compact and continue instead of stopping with a 1M-context credits error.
• Requests the API marks as not retryable are no longer retried, which could hold a turn for minutes, or indefinitely in unattended retry mode.

Later releases add more:

• claude -p now prints the error and exits with code 1 on a startup failure, instead of exiting 0 with no output. This matters for CI scripts (2.1.314).
• In VS Code and the Claude Desktop Code tab, a conversation over 200K tokens that sits idle is summarized, and the summary is used on your next message if the prompt cache has expired (2.1.314).
• Shell commands are no longer left running on macOS and Linux when Claude Code exits by itself, and hung network requests on truncated compressed responses are fixed (2.1.316).
• Per the CHANGELOG, 2.1.312 (October 27) fixes managed settings not loading after sign-in from VS Code, large CLAUDE.md files being overwritten, and skills not being found after a cd in Bash. It also adds concurrency limits for prompt and agent hooks and click-to-send for queued messages in fullscreen mode, and changes usage-limit wording to "5-hour limit". The claude_code.safety_event OpenTelemetry event is no longer sent.

Gateways and cloud providers

• Claude apps gateway (2.1.316): pooled group budgets (one monthly budget members share, set with pooled_budget on the spend limits admin API), a layered policy mode (managed.policy_mode: layered), and guardrail.input_tagging for Amazon Bedrock upstreams. Version 2.1.312 adds an oidc.user_id_claim setting, so users can be identified by an id_token claim other than sub.
• Model catalog (2.1.311): on Bedrock, Mantle, Vertex and Foundry, Claude Code loads model settings from a signed file on downloads.claude.ai. Set CLAUDE_CODE_MODEL_CATALOG=off to disable it, which matters for locked-down networks.
• Model names (2.1.312): on Bedrock, Vertex, Foundry and Mantle, unrecognized model names get a shorter system prompt, and automatic model fallback extends to those names.
• Fallbacks: the sonnet, opus and haiku aliases on Vertex AI and Bedrock switch to an older model instead of erroring when the newest is not available to your account or region (2.1.316). Microsoft Foundry falls back to an older model when no deployment exists for the default Sonnet, Opus or Haiku model, and Bedrock sessions with many images drop older images and compact (2.1.314).
• Cloud sessions (2.1.316): a form card lets you fill in or decline a form an MCP server asks for, and Enterprise members can add and delete network secrets on their own cloud environments.
• Other 2.1.311 changes: DISABLE_COMPACT in settings files can now only turn compaction off, not back on. The VS Code extension starts sessions with telemetry disabled when VS Code's own telemetry setting is off. The self-hosted runner gains --hold-idle-file and --max-hold-min to keep an idle session alive while the host reports leftover processes.

OpenAI Codex: 0.165.0 and 0.166.0

Both are on the Codex releases page.

0.166.0 (October 29) is the latest stable build. The notes on the page are truncated, so check the full notes before upgrading. What is listed:

• A keyboard-driven transcript selection with word, line and page movement plus search, and Markdown rendering in TUI questions and answers.
• GPT-6.1 Terra in the bundled, Vertex AI and Amazon Bedrock model catalogs.
• codex queue --resume to resume unloaded threads, and codex doctor --network-only for focused connectivity diagnostics.
• Fixes: approval and permission settings now persist across thread resume and fork, and reasoning-token accounting carries over on resume and fork.

0.165.0 (October 27) brought:

• A built-in Google Cloud Vertex AI provider, selected with model_provider = "gcp".
• GPT-6.1 Terra added to the model catalogs, a repository filter in the session picker, and reasoning effort that can change mid-conversation for supported models.
• Memory v2 as the default for fresh profiles, and attached thread forks in the app-server API as an experimental feature.
• Fixes: queued agent messages no longer restart work after an interrupt or a switch to Plan mode, retries stop when a provider reports a model as not found, and a marketplace URL validation gap is closed.

Research: Embedded Browsers in Five AI Coding Tools

According to the Help Net Security report, Forever Security researchers found that Cursor, GitHub Copilot, OpenAI Codex, VS Code and Claude Desktop each loosened their embedded browser so the AI could preview local projects or read tab content, and in all five one of the three core browser rules broke. They call the pattern EMBRACE. The reported findings and fix status:

• Cursor: a page with a hidden agent command could be run with one Enter press, and the agent disabled its own sandbox when blocked. Patched October 1.
• GitHub Copilot: the app trusted a page's reported address, so a fake page could show github.com/login over a fake sign-in form. Fixed in version 1.1.29 on October 13.
• OpenAI Codex: a single local file could read the whole disk, including SSH keys, .env files and npm tokens on the researchers' test machine. No fix as of October 29.
• VS Code: a web worker sits outside the project-folder check and can read any file. No fix as of October 29. Microsoft declined the report, citing Workspace Trust.
• Claude Desktop: the article describes the same folder-check gap as VS Code. It says Anthropic declined Forever's report but accepted a similar one from Noma Security, and does not state a fix.

This is one researcher's account reported by a news site, and the vendors' own responses were not in the article, so treat the details as unconfirmed until the vendors publish advisories.

What to Do

• Search your settings, scripts and CI for Opus 4.8 fast mode and move them to Opus 5.5 before November 30.
• Update Claude Code to 2.1.316, and consider enabling permissions.blockWritesOutsideWorkingDirectories in shared settings.
• If you run subagents heavily, set CLAUDE_CODE_SUBAGENT_EFFORT so they do not inherit a high session effort.
• Gateway admins: review layered policy mode and pooled budgets before rolling them out to groups.
• Try Fusion at Standard effort on a real task and compare cost and quality against your usual model before the discount ends.
• Pilot Agentic Workflows with read-only defaults before granting write outputs, and try assisted permissions in a throwaway repo before enabling them for a team.
• If you build internal agents on the Copilot SDK, trial server-to-server auth in a test org. It is a preview, so don't depend on it for production yet.
• Check Memory v2 behavior on fresh Codex profiles, and read the full 0.166.0 notes before upgrading.
• Codex and VS Code users: don't treat the Workspace Trust prompt as protection, review which projects are trusted, and be careful with links an assistant generates.
• Copilot and Cursor users: confirm you are on Copilot 1.1.29 or later and a Cursor build from after October 1.

Sources

• GitHub changelog: Copilot Fusion
• GitHub changelog: Agentic Workflows GA
• GitHub changelog: Agent Merge
• GitHub changelog: Copilot code review Max and approvals
• GitHub changelog: Assisted permissions GA
• GitHub changelog: GPT-6.1 Terra in Copilot
• GitHub changelog: Server-to-server authentication
• GitHub changelog: Microsoft Copilot SDK
• GitHub changelog: GitHub Enterprise includes Copilot
• GitHub changelog: Issue artifacts
• GitHub changelog: Security extension for the Copilot app
• GitHub changelog: Copilot Automations
• GitHub Copilot changelog
• Claude Code releases
• Claude Code v2.1.311
• Claude Code CHANGELOG.md
• OpenAI Codex releases
• Help Net Security: embedded browser research