OpenClaw v2026.2.14 Released: Major Security Hardening and 100+ Bug Fixes
OpenClaw releases v2026.2.14 with extensive security improvements, TUI stability enhancements, memory system optimizations, and 100+ bug fixes across channels, agents, and tools.
OpenClaw has released version 2026.2.14 with a major focus on security hardening, terminal UI (TUI) stability, and memory system performance. The release includes over 100 bug fixes and addresses multiple security vulnerabilities across the platform.
Major Security Improvements
The release includes extensive security hardening across multiple components:
Memory and Data Protection:
- LanceDB memory system now treats recalled memories as untrusted context, escaping injected memory text and adding explicit non-instruction framing to reduce memory-poisoning risk
- Auto-capture feature now requires explicit
autoCapture: trueopt-in (default disabled) to prevent automatic PII capture - QMD security improvements include rawKeyPrefix support for scope rules and legacy keyPrefix matching preservation
SSRF and Path Traversal Protection:
- Hardened media URL fetching against SSRF across multiple channels (Discord voice messages, Feishu, Tlon)
- Archive extraction now enforces entry/size limits to prevent resource exhaustion
- Apply_patch tool enforces workspace-root path bounds to block traversal and symlink escape writes
- Browser control file upload/download helpers now prevent path traversal and local file disclosure
Webhook and Authentication:
- Telegram webhooks now require non-empty
webhookSecretto prevent unauthenticated request forgery - Voice call services (Telnyx, Twilio) now enforce webhook signature verification
- BlueBubbles webhook auth hardened behind reverse proxies, requiring passwords for proxied requests
- Multiple channels now reject ambiguous shared-path webhook routing
Command Injection Prevention:
- Windows: Avoids shell invocation when spawning child processes to prevent cmd.exe metacharacter injection
- macOS: Prevents shell injection when writing Claude CLI keychain credentials
- Scripts: Validates GitHub logins and avoids shell injection in update-clawtributors.ts
TUI Stability and Performance
The terminal user interface received significant stability improvements:
Rendering Enhancements:
- Sanitizes ANSI/control-heavy history text and redacts binary-like lines to prevent startup crashes on binary attachment history
- Renders assistant body text in terminal default foreground for better contrast on light themes (Solarized Light)
- Preserves in-flight streaming replies when concurrent runs finalize
- Keeps pre-tool streamed text visible when tool-boundary deltas omit earlier text blocks
Session Management:
- Honors explicit
--sessionflag even whensession.scopeis global - Uses available terminal width for session name display in searchable select lists
- Fixes lazy core command registration for maintenance commands (doctor, dashboard, reset, uninstall)
Memory System Optimizations
QMD (Query Memory Database) and builtin memory backends received multiple performance and correctness improvements:
QMD Performance:
- Caps command output buffering to prevent memory exhaustion
- Parses QMD scope keys once per request
- Queries QMD index using exact docid matches before prefix lookup
- Passes result limits to search/vsearch commands for earlier result capping
- Skips rewriting unchanged session export markdown files to reduce disk churn
- Avoids reading full markdown files when from/lines window is requested
Multi-Collection Query Fix:
- Runs one
qmd query -cper managed collection and merges by best score to avoid ranking corruption - Detects null-byte ENOTDIR update failures, rebuilds collections once, and retries to self-heal corrupted metadata
Memory Leak Prevention:
- Bounds diagnostic session state, agentRunSeq tracking, ABORT_MEMORY, thread-starter cache, directory cache, and remote-skills cache growth
- Implements TTL + max-size pruning across multiple long-running gateway components
Channel and Plugin Improvements
New Features:
- Telegram: Poll sending support via
openclaw message pollwith duration, silent delivery, and anonymity controls - Discord: Exec approval prompts can now target channels or both DM+channel via
channels.discord.execApprovals.target - Slack/Discord: New
dmPolicyandallowFromconfig aliases for DM access control
Fixes:
- CLI/Plugins: Ensures
openclaw message sendexits after successful delivery and runs registeredgateway_stophooks before exit - WhatsApp: Honors per-account dmPolicy overrides
- LINE: Returns 200 OK for Developers Console βVerifyβ requests without X-Line-Signature
- Cron: Delivers text-only output directly when delivery.to is set; preserves agent identity in Slack outbound messages
Agent and Tool Enhancements
Workspace and Bootstrap:
- Persists bootstrap onboarding state for partially initialized workspaces
- Creates BOOTSTRAP.md when core workspace files are seeded
- Hardened transcript path resolution for mismatched agent context
Tool Improvements:
- Media tool accepts MEDIA:-prefixed paths with lenient whitespace
- Image tool allows workspace-local image paths by including active workspace in allowlists
- Sandbox file tools are now bind-mount aware with read-only bind semantics enforcement
Provider Resilience:
- Treats empty-stream provider failures as timeout-class failover signals
- Classifies external timeout aborts during compaction same as internal timeouts
- Adds safety timeout around embedded session.compact() to prevent stalled compaction blocking
Cron System Fixes
- Prevents cron list/status from silently skipping past-due recurring jobs
- Repairs missing/corrupt nextRunAtMs without globally recomputing unrelated jobs
- Skips missed-job replay on startup for jobs interrupted mid-run, preventing restart loops
Breaking Changes and Deprecations
Security-Related:
- Telegram allowlist authorization now requires numeric sender IDs (rejects @username principals)
- Google Chat users/ allowlists deprecated (treat users/β¦ as immutable user id only)
- Hooks transformsDir must now be within ~/.openclaw/hooks/transforms
Configuration:
- Auto-capture for LanceDB memory now defaults to disabled (requires explicit opt-in)
Community Contributions
This release includes contributions from over 80 community members, with special thanks to:
- @robbyczgw-cla (multiple improvements)
- @vincentkoc (extensive security hardening)
- @p80n-sec (security improvements)
- @christos-eth (security fixes)
- @coygeek and @vignesh07 (memory leak fixes)
Upgrade Recommendations
Users are strongly encouraged to upgrade to v2026.2.14 due to the extensive security improvements, particularly:
- Webhook authentication hardening
- SSRF and path traversal protections
- Memory poisoning risk reduction
- Command injection prevention
Source: OpenClaw GitHub Releases
Related Articles
OpenClaw v2026.2.15 Released: Discord Components v2 and Nested Sub-Agents
OpenClaw releases v2026.2.15 one day after v2026.2.14, adding Discord interactive components, nested sub-agents with configurable depth, plugin hook enhancements, and continued security hardening.
Kimi Integrates OpenClaw Natively - 5,000+ Community Skills and 40GB Cloud Storage in Browser
Moonshot AI's Kimi.com now supports OpenClaw natively in browser tabs, offering 24/7 uptime, ClawHub access with 5,000+ skills, 40GB cloud storage, and pro-grade search capabilities.
Cursor 2.5 Brings Plugin Marketplace, Sandbox Network Controls, and Async Subagents
Cursor releases version 2.5 with a plugin marketplace for extensibility, granular network access controls for sandboxed environments, and asynchronous subagent execution for parallel processing.
Popular Articles
Claude Code v2.1.93 Released - Deferred Permission Decisions, Flicker-Free Rendering, and More
Anthropic releases Claude Code v2.1.93 with deferred permission decisions for PreToolUse hooks, flicker-free rendering option, PermissionDenied hook, and named subagent typeahead support.
Claude Code v2.1.92 Released - forceRemoteSettingsRefresh, Bedrock Setup Wizard, and More
Anthropic releases Claude Code v2.1.92 with forceRemoteSettingsRefresh policy setting, AWS Bedrock setup wizard, /cost command improvements, and numerous bug fixes.
Claude Code v2.1.84 Release - PowerShell Tool Preview and Environment Configuration Enhancements
Claude Code v2.1.84 introduces PowerShell tool for Windows, new environment variable overrides for model selection, idle session handling improvements, and various stability fixes.
Latest Articles
Claude Code v2.1.93 Released - Deferred Permission Decisions, Flicker-Free Rendering, and More
Anthropic releases Claude Code v2.1.93 with deferred permission decisions for PreToolUse hooks, flicker-free rendering option, PermissionDenied hook, and named subagent typeahead support.
Claude Code v2.1.92 Released - forceRemoteSettingsRefresh, Bedrock Setup Wizard, and More
Anthropic releases Claude Code v2.1.92 with forceRemoteSettingsRefresh policy setting, AWS Bedrock setup wizard, /cost command improvements, and numerous bug fixes.
Claude Code v2.1.91 Released - MCP Tool Result Persistence and Improved Edit Tool
Claude Code v2.1.91 introduces MCP tool result persistence override, improved shell execution controls, and enhanced Edit tool efficiency.