OpenClaw 12 min read

Is OpenClaw Dead? The Release Data Says No

Is OpenClaw dead? No. 2026.11.2 shipped 8 Nov and 2026.10.10 had 3,028 PRs from 454 contributors. The release data, and why it still feels dead.

Shabnam Katoch

Shabnam Katoch

Growth Head

Is OpenClaw Dead? The Release Data Says No

A Medium post declaring "OpenClaw Is Dead" went viral in May 2026. Forrester published "OpenClaw Is Dead, Long Live OpenClaw." Palo Alto Networks called it a "security nightmare." Here's the honest assessment from a team that tracks the OpenClaw ecosystem every day, refreshed with the release data as of 13 October 2026.

In May, Mehul Gupta published "OpenClaw Is Dead" on Medium's Data Science in Your Pocket. The thesis: OpenClaw's security disasters, ecosystem chaos, and branding instability have killed the project's reputation even though the code is still active.

A month before that, Forrester published "OpenClaw Is Dead, Long Live OpenClaw" after Anthropic's April 4 ban forced every Claude Pro/Max user off third-party tools. One analyst documented his bill jumping from $20/month flat to $25-50/day overnight.

And before that, Palo Alto Networks called OpenClaw a "security nightmare." CrowdStrike published a full enterprise security advisory. Cisco found a skill exfiltrating data without user awareness. The OpenClaw maintainer himself warned: "if you can't understand how to run a command line, this is far too dangerous of a project for you to use safely."

So. Is OpenClaw dead?

No. But the question is worth taking seriously.

Is OpenClaw dead? The October 2026 numbers

Here is what landed on OpenClaw's GitHub releases page recently:

  • 2026.11.2 (8 November): the current stable release, marked Latest. What changed and how to update: OpenClaw latest version.
  • 2026.11.1 (5 November): channel reply and update recovery fixes.
  • 2026.10.10 (3 November): the biggest release since 2.0. 4,825 commits, 3,028 pull requests and 454 contributors.
  • 2026.10.4 to 2026.10.9 (16 to 29 October): six stable releases in under two weeks, plus extended-stable 2026.9.33 to 2026.9.36.
  • 2026.10.3 (13 October): a fix release for updates that would not go through.
  • 2026.10.2 (12 October): a cleanup release for upgrades from 2026.10.1.
  • 2026.10.1 (10 October): the big one. 3,055 commits, 2,425 pull requests and 402 contributors, per its release header.
  • 2026.9.9 (8 October): a stable patch release two days earlier (2026.9.8 shipped on 3 October).
  • 2026.10.5-beta.1 (11 October, pre-release): Claude Haiku 5.5 support, a new X (Twitter) channel plugin, browsing and searching ClawHub plugins and skills from inside the Gateway, and a native worker inference runtime for running inference locally. Two more betas followed on 12 and 13 October.
  • Extended-stable 2026.8.34 and 2026.8.35 (2 October): OpenClaw calls extended-stable "our current equivalent to LTS." 2026.8.34 backports 113 audit-selected fixes across upgrades, authentication, sessions, channels, plugins and sandboxing; 2026.8.35 adds further security and ownership hardening.
  • Tencent joins ClawHub's skill review (2 October): OpenClaw integrated Tencent's AI-Infra-Guard (AIG) into ClawScan, the open-source tool behind ClawHub's security review. Every uploaded skill and plugin is now scanned by AIG and NVIDIA's SkillSpector independently, and an AI judge makes the final call (announcement).

That is six non-beta releases between 2 and 12 October, counting the two extended-stable builds. The pace has picked up since September, when four releases ran from 2026.8.1 to 2026.9.2 (see our 2026.9.2 breakdown).

Where the "dead" feeling comes from

If the code is this active, why does "is OpenClaw dead?" keep getting searched? Three things make the project look worse than its commit log.

A wall of "Update failure" issues. Between 1 and 13 October, more than 260 issues titled "Update failure: …" were opened on the OpenClaw GitHub repo. They share one template, such as "Update failure: activating (2026.10.1)", because OpenClaw's updater offers to file a sanitized report when an update fails and the user confirms. Several carry P0 release-blocker labels. Skim the issue list and it looks like a project on fire. Read it closely and it is a lot of people on a fast release train, many of them upgrading from older versions. If you hit one yourself, our decoder entry for update failure: reconcile abandoned explains what the verdict means.

The April Anthropic subscription episode. When Anthropic cut Claude Pro and Max subscriptions off from third-party tools on April 4, a lot of users left and never came back, even after the reversal on May 13. That exodus is where much of the "dead" talk started. The full timeline is in our Claude subscription and OpenClaw post.

A Mac app that drops older Macs. OpenClaw's docs now say the native OpenClaw.app requires macOS 26.2 (Tahoe) or later. The CLI and Gateway still run on macOS 13.5 and up, but anyone on an older Mac who downloads the app finds it simply won't run, which reads as "broken" rather than "moved on."

The case for "dead" (what the critics get right)

Six events fueling the OpenClaw is dead narrative: ClawHavoc, CVE-2026-25253, exposed instances, Anthropic ban, rough week, Meta incident

The Medium post isn't wrong about the facts. Here's what actually happened:

ClawHavoc (January 2026). 1,400+ malicious skills discovered on ClawHub. Typosquatted names distributing Atomic Stealer malware. SSH keys, API tokens, and browser cookies exfiltrated via reverse shells. ClawHub removed 2,419 suspicious skills. A Snyk audit found 13.4% of all skills had critical security issues.

CVE-2026-25253 (January 29). A one-click remote code execution vulnerability. CVSS 8.8. Patched the same day, but the exposure window was real.

500K+ exposed instances. Censys, Bitsight, and Hunt.io found 30,000+ OpenClaw instances on the public internet without authentication. The latest count: 500K+. These are agents with full system access, reachable by anyone.

The Anthropic ban (April 4). Anthropic banned Claude Pro and Max subscriptions from third-party tools including OpenClaw. Overnight, users went from $20/month flat to per-token API billing. The Forrester analyst documented $25-50/day costs. The community was furious. (Anthropic reversed the ban on May 13 and paused its planned replacement credit system on June 15.)

The "rough week" (April 29). OpenClaw officially apologized for v2026.4.29 breaking gateways, plugin dependency repair loops, and channel degradation. The blog post said: "That sucks. I'm sorry."

The Meta incident. Researcher Summer Yue's OpenClaw agent mass-deleted 200+ emails while ignoring stop commands. Meta subsequently banned OpenClaw on work devices.

The honest assessment: The security concerns are real. The stability concerns are real. The cost concerns are real. Anyone who dismisses these as FUD isn't paying attention.

For the complete security analysis of OpenClaw in 2026, our guide covers every CVE, every advisory, and the specific mitigations.

The case for "alive" (what the critics miss)

Here's what the "OpenClaw Is Dead" narrative gets wrong.

Six counterpoints showing OpenClaw is alive: GitHub stars, npm downloads, May releases, contributors, plugin externalization, LTS announcement

230K+ GitHub stars. Still growing. Not declining.

1.27M weekly npm downloads. That's not a dying project. That's the most-installed AI agent framework in existence.

15 releases in May 2026. v2026.5.1 through v2026.5.18 in 19 days. Including plugin externalization (v2026.5.12), voice agent improvements (v2026.5.4), typed plugin SDK, Docker security hardening, and a full Mac app redesign.

850+ contributors. Active PRs. Active issue triage. Peter Steinberger joined OpenAI, but the project moved to an open-source foundation and development accelerated, not slowed.

The architectural maturation is real. v2026.5.12 externalized WhatsApp, Slack, Bedrock, and Vertex from core. This is the kind of modular restructuring that mature projects do. Not dying projects.

An LTS-style channel now exists. The "rough week" blog promised an LTS release "later in May." It arrived as the extended-stable channel, which OpenClaw calls "our current equivalent to LTS": the 2026.8.34 and 2026.8.35 builds on 2 October 2026 backported audited fixes and security hardening for teams that don't want to ride every release.

The honest assessment: OpenClaw is not dead. But it's in a dangerous period where its growth has outpaced its security maturity. The project's response (VirusTotal partnership, plugin externalization, security hardening) suggests the maintainers understand the problem. Whether they can solve it fast enough to keep enterprise trust is the real question.

What the "is it dead" question actually means (the real issue)

Four kinds of trust at stake: skills trust, update trust, security trust, and cost trust

Here's where most people get it wrong.

"Is OpenClaw dead?" is not a question about the code. The code is clearly alive. 15 releases in May. The question is about something else entirely.

It's a question about trust.

  • Can you trust the ClawHub skill marketplace? (7.6% malicious rate says: not without vetting.)
  • Can you trust the update process? (v2026.5.5 broke GPT-5.5 configs. v2026.5.6 hotfixed it the same day.)
  • Can you trust the security model? (500K+ exposed instances says: not with default settings.)
  • Can you trust the cost model? ($178/week unoptimized says: not without manual optimization.)

The people saying "OpenClaw is dead" are really saying "I don't trust OpenClaw anymore." And trust, once lost, is harder to rebuild than code.

If the trust question is what concerns you, and you want an AI agent that is persistent, always-on, and multi-channel without the security, stability, and cost uncertainties of self-hosting, that is the gap a managed AI agent platform like BetterClaw fills. BetterClaw runs your agent in the cloud with Telegram, Slack, Discord and Gmail built in, plus webhooks, 500+ connectors, unlimited skills and 28+ model providers via your own API key with no markup. Free plan: 1 agent, 100 credits a month, 3 connectors, no card.

What actually happens next (three scenarios)

Three scenarios for OpenClaw's future: Linux of AI agents, market fragmentation, catastrophic incident

Scenario 1 (most likely): OpenClaw becomes the Linux of AI agents. Powerful, complex, expert-friendly. Enterprises adopt cautiously via extended-stable and security wrappers like NemoClaw, NVIDIA's wrapper around OpenClaw; casual users move to managed platforms.

Scenario 2 (already happening): The market fragments. OpenClaw keeps the developer core, while Hermes, Manus, NemoClaw and managed platforms (BetterClaw, Blink Claw, xCloud) each take a slice. The "one framework to rule them all" story ends.

Scenario 3 (unlikely but possible): A catastrophic incident. An exposed instance causes real financial harm or a headline breach, enterprise adoption freezes, and the code survives with permanent reputational damage.

If you're weighing a move off self-hosting, our OpenClaw alternative page covers the specific trade-offs between self-hosted and managed approaches.

The perspective from inside the ecosystem

Here's what we see tracking OpenClaw's ecosystem every day.

The agent model works. Persistent agents on messaging channels with tool access, memory, and autonomous task execution. That's not hype. That's real value for real users. The model works. The implementation has security debt.

The security debt is being addressed, slowly. Plugin externalization (v2026.5.12). VirusTotal scanning on ClawHub. Docker security hardening (NET_RAW/NET_ADMIN dropped). The typed plugin SDK. These are real improvements. They're also 6 months late compared to where the security maturity should be for a project with 500K+ exposed instances.

The update pace is both a strength and a weakness. 15 releases in 19 days means bugs get fixed fast. It also means bugs get introduced fast. v2026.5.5 broke GPT-5.5 configs. v2026.5.6 fixed it the same day. That's impressive response time. It's also evidence that the QA process can't keep up with the release velocity.

The real question isn't "is OpenClaw dead?" It's "should I run OpenClaw myself, or should I use a platform that manages the complexity for me?" That question has a different answer for different people.

For the best OpenClaw skills to install safely, our curated guide covers the verification process.

The honest take

Here's the perspective I wish the Medium post had included.

OpenClaw is not dead. It's maturing. Painfully, publicly, and with real casualties (stolen credentials, broken configs, spiraling costs). But maturation is not death. Node.js went through a similar phase. Docker went through it. Kubernetes went through it. Every infrastructure project that grows faster than its security story has this reckoning.

The question for you is personal. Do you have the time, expertise, and risk tolerance to manage a self-hosted OpenClaw instance with proper security? If yes, OpenClaw remains the most capable open-source agent framework available. If no, the managed platforms exist specifically because this question has an honest answer for most people: no.

We built BetterClaw because we watched the security incidents stack up and realized that "you can fix this yourself" is not a satisfactory answer for the 90% of users who don't want to become OpenClaw security engineers. The agent capabilities are real. The infrastructure burden shouldn't be.

If you want those capabilities without the security lottery, the update anxiety, and the cost surprises, give BetterClaw a try. BetterClaw runs your agent in the cloud with Telegram, Slack, Discord and Gmail built in, plus webhooks, 500+ connectors, unlimited skills and 28+ model providers via your own API key with no markup. Free plan: 1 agent, 100 credits a month, 3 connectors, no card. OpenClaw isn't dead. But you don't have to self-host anything to get an always-on agent.

Frequently Asked Questions

Is OpenClaw dead in 2026?

No. As of 10 November 2026, the current stable release is 2026.11.2 (8 November), and 2026.10.10 (3 November) shipped 4,825 commits and 3,028 pull requests from 454 contributors. OpenClaw also shipped 15 releases in the first 19 days of May 2026. The "OpenClaw Is Dead" narrative refers to reputation and trust, not code activity. The security concerns (ClawHavoc, CVE-2026-25253, 500K+ exposed instances) are real but are being addressed through plugin externalization, VirusTotal scanning, and security hardening.

Why did the "OpenClaw Is Dead" Medium post go viral?

Because it named real problems that users experience: 1,400+ malicious skills on ClawHub, a one-click RCE vulnerability (CVSS 8.8), the Anthropic April 4 ban forcing users from $20/month flat to $25-50/day API billing, the Meta email deletion incident, and the "rough week" where v2026.4.29 broke gateways and channels. The post resonated because the frustrations are genuine.

Is OpenClaw still maintained?

Yes. OpenClaw shipped more than a dozen non-beta releases between 13 October and 8 November 2026, ending with 2026.11.2 (the current stable release). Earlier, between 2 and 13 October, it shipped seven, including 2026.10.3, 2026.10.2, 2026.10.1 and the extended-stable 2026.8.34 and 2026.8.35 builds, plus 2026.10.5 betas on 11, 12 and 13 October. Issues are actively triaged and labelled on GitHub, and an extended-stable channel (OpenClaw's equivalent to LTS) gets backported fixes.

Is OpenClaw safe to use in 2026?

With proper configuration: yes, for technical users. Default settings are not safe for production. Disable public access, enable authentication, vet every ClawHub skill before installing, keep versions updated, and keep exec approvals tight; if commands start getting blocked, our decoder entry for exec denied: allowlist miss explains why and how to fix it safely. ClawHub's review got stronger on 2 October 2026, when Tencent's AI-Infra-Guard joined NVIDIA's SkillSpector in ClawScan. For non-technical users or enterprises, a managed platform handles security at the platform level.

What replaced OpenClaw?

Nothing has replaced it; OpenClaw is still under very active development. What has changed is that many users no longer self-host it. Developers who want a different framework look at Hermes Agent or NanoClaw, and people who just want an always-on agent move to a managed platform. Our OpenClaw alternative page compares the trade-offs.

What are the alternatives if I'm worried about OpenClaw security?

Managed platforms: BetterClaw (Telegram, Slack, Discord and Gmail built in, unlimited skills, Free plan with 1 agent, 100 credits a month and 3 connectors), Blink Claw, xCloud. Security wrappers: NemoClaw (NVIDIA's enterprise wrapper). Alternative frameworks: Hermes Agent (self-improving, reportedly more stable), NanoClaw (15 files, minimal attack surface), ZeroClaw (Rust-based, minimal). Each trades some of OpenClaw's breadth for better security or simplicity.

Should enterprises use OpenClaw in 2026?

CrowdStrike published a full security advisory recommending against unmanaged OpenClaw in enterprise environments. For enterprises considering AI agents, BetterClaw offers a free AI readiness audit that identifies the highest-impact use cases, shares a clear proposal with expected ROI, and implements on the BetterClaw platform if it makes sense. No commitment required.

Want to skip the setup?

BetterClaw does this in 60 seconds. No Docker, no config files.

Start free
Tags:is OpenClaw deadOpenClaw security crisisOpenClaw dead MediumOpenClaw 2026 problemsOpenClaw alternativeOpenClaw securityOpenClaw reputation
Shabnam Katoch

Written by

Shabnam Katoch

Growth Head

Shabnam leads growth at BetterClaw and has spent the past three years building organic distribution for AI-native products. She writes about no-code agent builders, content strategy, and what it actually takes to grow without an ads budget.

Share this article
Was this helpful?