Free reference

Claude Code and Codex: the changes that touch your controls.

Both agents ship several releases a week, and most release notes are about the interface. This page keeps only the lines that matter if you own the policy: managed settings, permissions, sandboxing, network rules, MCP, hooks, telemetry, auth and security fixes. Each row links the release note it came from.

Last checked .

How to read it

  • Each row is a one-sentence summary of one release-note line. Follow the version link and read the vendor's wording before you change a policy.
  • The selection is a judgment call and it is not complete. The release notes were filtered by keyword, so a relevant change with unusual wording can be missing.
  • Claude Code's changelog carries no dates. The dates shown for it are the npm publish dates of each version. Codex dates are the GitHub release dates.
  • Only stable Codex releases are covered. Nothing here was tested on a machine; it reports what the release notes say.
  • I sell The Coding Agent SOC 2 Kit, which builds managed policy files for both agents. This page is free and does not need it.

Claude Code

54 changes, read through version 2.1.322.

Version What changed Area Date
2.1.321 Fixed a case where a read-only Bash command ran outside the sandbox without a permission prompt. Security fix
2.1.321 Deny rules and protected-file prompts now apply to the file written by a cp or mv into a directory. Security fix
2.1.319 Fixed a Bash permission bypass where a crafted environment prefix hid part of a command from permission checks. Security fix
2.1.319 Fixed a sandbox escape that occurred when creating a worktree. Security fix
2.1.319 Fixed a case where a sed command could skip the permission check in accept-edits and auto modes. Security fix
2.1.318 deniedMcpServers URL entries now also block matching Anthropic-provided servers and cloud or self-hosted-runner connectors. MCP
2.1.316 Server-managed settings for Team and Enterprise logins apply without an approval prompt, and a stored copy that fails verification is refused. Managed settings
2.1.314 Fixed forceRemoteSettingsRefresh set only in remote managed settings, which let sessions start without policy after the stored copy expired. Managed settings
2.1.314 Fixed sandbox protections not applying to a worktree entered through claude -p with --resume or --continue. Sandbox
2.1.312 Claude Code stopped sending the claude_code.safety_event OpenTelemetry event. Telemetry
2.1.309 Fixed sandboxed Bash commands running the shell startup files outside the sandbox before each command. Sandbox
2.1.309 Fixed sandboxed commands and auto-accepted edits being able to modify the MCP config file passed with --mcp-config. Sandbox
2.1.307 Added the claude_code.safety_event OpenTelemetry event for tool calls stopped, prompted or sent back by rules, auto mode, hooks or the sandbox. Telemetry
2.1.305 Added sandbox.allowNestedUserNamespaces, which lets sandboxed Linux commands create user namespaces and thereby get around the sandbox's file protections. Sandbox
2.1.305 allowedMcpServers no longer filters managed-mcp.json stdio servers that use variables only in env, so servers the allowlist omitted now load; deniedMcpServers keeps them off. MCP
2.1.299 Improved enforcement of managed permission rules, managed hooks and organization MCP tool names against user-installed mods. Hooks
2.1.299 MCP allow rules now apply only to the server they name, so mcp__a no longer allows servers named a__b or a_. MCP
2.1.299 Fixed a sandboxed shell command being able to write into a --plugin-dir folder whose code runs outside the sandbox. Sandbox
2.1.296 Fixed handling of managed PreToolUse hooks that deny with continue false and managed prompt hooks that block a tool call. Hooks
2.1.292 Security fix: PreToolUse hook approvals and auto mode no longer bypass the permission prompt for file reads from network (UNC) paths. Security fix
2.1.292 Fixed a tampered on-disk cache of server-managed settings being able to switch off the built-in policy plugin while the settings fetch failed. Managed settings
2.1.290 Added a warning when a managed settings file is a link to a file outside the managed settings folder. Managed settings
2.1.288 Fixed a dangerous rm inside a bash -c or sh -c script running without a prompt in bypassPermissions mode or under a shell allow rule. Permissions
2.1.285 Project settings can no longer widen or turn off an admin-required sandbox, replace a managed proxy, extend a strict allowlist, or reopen managed read-denies. Sandbox
2.1.285 Claude Code now warns and starts without policies when the OS denies reading the managed settings file; other read errors and unparseable files stop every session. Managed settings
2.1.284 Fixed marketplace, claude.ai and npm plugins pre-approving their own tools under managed allowManagedPermissionRulesOnly. Permissions
2.1.283 Managed sandbox settings with one invalid nested value now fail closed, and the rest of the sandbox block still applies. Sandbox
2.1.282 sandbox.excludedCommands entries from project and local settings are ignored when managed settings forbid unsandboxed commands or restrict managed domains. Sandbox
2.1.282 Project and local settings now ignore OpenTelemetry variables that turn on export, set the endpoint or capture content. Telemetry
2.1.282 On Windows and WSL, an invalid or unreadable admin policy now keeps user-writable HKCU and WSL settings from applying. Managed settings
2.1.282 Added the allowClaudeInChromeWithManagedMcp managed setting so claude --chrome can run alongside an exclusive managed-mcp.json. MCP
2.1.277 Fixed a sandbox.excludedCommands glob exempting a whole compound Bash command from the sandbox when only one part matched. Sandbox
2.1.274 Added the claude_code.managed_settings_resolved OTel event listing managed-settings sources and policy helper state, with redacted settings and digests optional. Telemetry
2.1.273 Fixed allowManagedMcpServersOnly, deniedMcpServers and disableClaudeAiConnectors from MDM or managed-settings.json being ignored when server-managed settings were also present. MCP
2.1.271 An unreadable or unparseable enterprise managed-mcp.json now keeps exclusive MCP control and warns at startup instead of being ignored. MCP
2.1.267 Managed allowedHttpHookUrls, httpHookAllowedEnvVars and allowedChannelPlugins now admit nothing, rather than everything, when unreadable. Hooks
2.1.259 Claude Code now refuses to start and names the source when managed settings from a file, drop-in, MDM plist or HKLM value cannot be parsed. Managed settings
2.1.257 defaultMode bypassPermissions in project settings files is now ignored; it must be set in user or managed settings or on the command line. Permissions
2.1.257 Added a Containment Escape rule to auto mode so cloud metadata-credential fetches, egress evasion and cross-tenant reach are no longer auto-approved by default. Permissions
2.1.251 Fixed project settings enabling detailed beta tracing or raw API body logging, and lower-scope tracing endpoints bypassing a managed OTLP collector. Telemetry
2.1.251 Server-managed settings that terminate sandbox TLS, route sandbox traffic through a proxy, inject credentials or weaken isolation now require approval. Managed settings
2.1.248 Added --restricted, which removes command-running tools and WebFetch, confines file tools to the working directory, and refuses bypassPermissions. Permissions
2.1.232 Hardened the Linux filesystem sandbox against a protected-path bypass. Security fix
2.1.232 Fixed a PowerShell permission bypass where variable-writing parameters could overwrite $PSDefaultParameterValues and redirect file access. Security fix
2.1.223 Fixed an agent definition's bypassPermissions mode ignoring the organization policy that disables bypass-permissions. Permissions
2.1.221 Fixed a Bash permission-check bypass where zsh regex conditionals could run hidden commands; affected commands now prompt. Security fix
2.1.219 Added sandbox.network.strictAllowlist to deny non-allowlisted hosts for sandboxed commands without prompting. Network
2.1.212 Enterprise forceLoginMethod is now enforced for VS Code extension, SDK, setup-token and install-github-app logins, not only the terminal. Auth
2.1.193 Added the claude_code.assistant_response OpenTelemetry event with model response text, redacted unless OTEL_LOG_ASSISTANT_RESPONSES=1. Telemetry
2.1.187 Added the sandbox.credentials setting to block sandboxed commands from reading credential files and secret environment variables. Sandbox
2.1.175 Added the enforceAvailableModels managed setting so the availableModels allowlist also constrains the default model and cannot be widened by user or project settings. Managed settings
2.1.169 Fixed enterprise managed MCP policies not being enforced on reconnect, IDE-typed configs, --mcp-config servers, or before remote settings loaded. MCP
2.1.163 Added requiredMinimumVersion and requiredMaximumVersion managed settings; Claude Code refuses to start outside the allowed version range. Managed settings
2.1.149 Added the allowAllClaudeAiMcps managed setting to load claude.ai cloud MCP connectors alongside managed-mcp.json. MCP

Codex

54 changes, read through version 0.167.0.

Version What changed Area Date
0.167.0 The app-server README documents managed network constraints under permissions.network and prompt-user command rules that require fresh human approval. Network
0.166.0 Managed requirements can now carry OpenTelemetry configuration. Telemetry
0.166.0 Managed requirements gained support for a network_proxy table. Network
0.166.0 OTEL exporter headers are redacted in app-server config responses. Telemetry
0.165.0 Added managed configuration loaders for custom config servers. Managed settings
0.165.0 Added support for managed hooks scoped to an environment. Hooks
0.165.0 Added managed network review domains that are evaluated before allow rules. Network
0.164.0 Windows registry overrides are now supported for requirements.toml. Managed settings
0.164.0 Managed configuration is now required from the configured server. Managed settings
0.164.0 Enterprise MCP connections authenticate through eligible Codex account grants or a dedicated sign-in, including OAuth clients that require secrets. MCP
0.164.0 Credentials are masked during supported remote Bash commands and Zsh snapshot sessions, including credentials from custom providers. Auth
0.164.0 Hardened sandbox checks for case-variant metadata paths on macOS and for replaceable symlinks used to locate executables. Security fix
0.163.0 The network proxy checks permission before DNS lookups, rejects duplicate Host headers, and stops credential aliases from bypassing inspection hooks. Network
0.163.0 Resuming a saved thread restores its permissions and network approvals while respecting explicit overrides and current managed policy. Permissions
0.162.0 Fixed Linux sandbox startup with multiple denied files, rejected writable sandbox-construction executables, and stopped ripgrep configuration from weakening deny-glob masks. Sandbox
0.161.0 Approved filesystem escalation can grant broader write access while keeping denied reads and network restrictions, and background tasks keep their originating turn's permissions. Permissions
0.161.0 Managed requirements can now disable the Windows MXC sandbox. Sandbox
0.161.0 Authentication guidance now covers keyring storage instead of implying credentials always reside in auth.json. Auth
0.158.0 Terminal input approval is on by default for commands running with elevated permissions. Permissions
0.158.0 Direct exec-server WebSocket connections can be secured with bearer tokens, including those configured through app-server. Auth
0.157.0 Network restrictions are enforced across redirects and ongoing HTTP and WebSocket traffic, including cancellation when policy changes revoke access. Network
0.156.0 Closed sandbox isolation gaps involving inbound Windows connections, privileged Linux and macOS sockets, and writes through read-only macOS file handles. Security fix
0.156.0 Managed configuration now requires forced macOS preferences. Managed settings
0.155.0 Blocked Windows-process escapes from restricted WSL sandboxes and hardened brokered shell snapshots against credential exposure. Security fix
0.154.0 Startup no longer runs workspace-controlled helpers before trust is established, and the macOS sandbox blocks terminal input injection. Security fix
0.152.0 Cloud task requests reject untrusted backend URLs and disable redirects to protect saved credentials. Security fix
0.150.0 Untrusted projects no longer supply project-level AGENTS.md instructions, and managed deny-read rules stay enforced after permission changes. Permissions
0.150.0 Managed requirements can now include developer instructions. Managed settings
0.149.0 Documented DNS exfiltration risks and trust limitations for secure devcontainers. Network
0.148.0 Sandbox restrictions now fail closed for denied or unreadable paths on Linux and Windows. Sandbox
0.148.0 Sessions are rejected when required managed hooks cannot be loaded. Hooks
0.147.0 Unfamiliar local projects require explicit trust, and managed authentication restrictions are enforced before credentials are used. Auth
0.147.0 Plugin isolation was hardened and network access is denied when policy updates fail. Network
0.147.0 Secrets and complete bearer tokens are redacted from displayed commands and replayed conversation history. Auth
0.147.0 Added the --approve-for-me CLI flag to enable automatically reviewed approvals. Permissions
0.147.0 Removed the deprecated codex exec --full-auto flag; --sandbox workspace-write replaces it. Permissions
0.146.0 Exact values from managed config requirements are now enforced. Managed settings
0.146.0 Managed MCP requirements now refresh for active threads. MCP
0.146.0 Added a managed policy that controls in-app updates. Managed settings
0.145.0 Strengthened safety and approval handling with better forced-rm detection, consistent full-access confirmation, and preserved rejection reasons across tools. Permissions
0.143.0 Plugin marketplace source admission requirements are now enforced. Managed settings
0.143.0 Updated OpenSSL, Hono, fast-uri, quick-xml and crossbeam-epoch to address security advisories. Security fix
0.142.2 PowerShell commands with executable AST regions the safety classifier cannot inspect now require approval. Security fix
0.141.0 Remote executors use authenticated, end-to-end encrypted Noise relay channels. Auth
0.141.0 Hook trust bypass now persists through codex exec thread start and resume, and blocking PostToolUse hooks reject code-mode tool calls. Hooks
0.140.0 Added managed Amazon Bedrock API-key authentication and encrypted local storage for CLI and MCP OAuth credentials. Auth
0.139.0 Thread resets no longer drop cloud-managed requirements or feature flags during TUI config reloads. Managed settings
0.138.0 Managed permission profile allowlists are now enforced. Permissions
0.137.0 Enterprise and admin flows show monthly credit limits and can apply cloud-managed config bundles, including for EDU workspaces. Managed settings
0.136.0 Command-safety hardening stops /diff from running repository-provided Git helpers or hooks and rejects browser-origin exec-server websocket handshakes. Security fix
0.136.0 Windows admins get an alpha elevated sandbox setup command, plus requirements support for allowed Windows sandbox implementations. Sandbox
0.136.0 Deny read rules stay enforced for safe-command and approval-bypass paths. Permissions
0.133.0 Managed permission profiles are now supported in requirements.toml. Managed settings
0.131.0 Managed read restrictions are preserved during permission escalation. Permissions