OpenClaw has fastened a high-severity security situation that, if efficiently exploited, might have allowed a malicious web site to connect with a regionally working synthetic intelligence (AI) agent and take over management.
“Our vulnerability lives within the core system itself – no plugins, no market, no user-installed extensions – simply the naked OpenClaw gateway, working precisely as documented,” Oasis Safety stated in a report printed this week.
The flaw has been codenamed ClawJacked by the cybersecurity firm.
The assault assumes the next menace mannequin: A developer has OpenClaw arrange and working on their laptop computer, with its gateway, a neighborhood WebSocket server, sure to localhost and guarded by a password. The assault kicks in when the developer lands on an attacker-controlled web site by social engineering or another means.
The an infection sequence then follows the steps beneath –
- Malicious JavaScript on the net web page opens a WebSocket connection to localhost on the OpenClaw gateway port.
- The script brute-forces the gateway password by profiting from a lacking rate-limiting mechanism.
- Submit profitable authentication with admin-level permissions, the script stealthily registers as a trusted gadget, which is auto-approved by the gateway with none consumer immediate.
- The attacker beneficial properties full management over the AI agent, permitting them to work together with it, dump configuration knowledge, enumerate related nodes, and skim software logs.
“Any web site you go to can open one to your localhost. Not like common HTTP requests, the browser does not block these cross-origin connections,” Oasis Safety stated. “So whilst you’re searching any web site, JavaScript working on that web page can silently open a connection to your native OpenClaw gateway. The consumer sees nothing.”
“That misplaced belief has actual penalties. The gateway relaxes a number of security mechanisms for native connections – together with silently approving new gadget registrations with out prompting the consumer. Usually, when a brand new gadget connects, the consumer should affirm the pairing. From localhost, it is computerized.”
Following accountable disclosure, OpenClaw pushed a repair in lower than 24 hours with model 2026.2.25 launched on February 26, 2026. Customers are suggested to use the newest updates as quickly as attainable, periodically audit entry granted to AI brokers, and implement acceptable governance controls for non-human (aka agentic) identities.
The event comes amid a broader security scrutiny of the OpenClaw ecosystem, primarily stemming from the truth that AI brokers maintain entrenched entry to disparate techniques and the authority to execute duties throughout enterprise instruments, resulting in a considerably bigger blast radius ought to they be compromised.
Reviews from Bitsight and NeuralTrust have detailed how OpenClaw cases left related to the web pose an expanded assault floor, with every built-in service additional broadening the blast radius and may be remodeled into an assault weapon by embedding immediate injections in content material (e.g., an e mail or a Slack message) processed by the agent to execute malicious actions.
The disclosure comes as OpenClaw additionally patched a log poisoning vulnerability that allowed attackers to jot down malicious content material to log recordsdata by way of WebSocket requests to a publicly accessible occasion on TCP port 18789.
Because the agent reads its personal logs to troubleshoot sure duties, the security loophole could possibly be abused by a menace actor to embed oblique immediate injections, resulting in unintended penalties. The difficulty was addressed in model 2026.2.13, which was shipped on February 14, 2026.
“If the injected textual content is interpreted as significant operational info slightly than untrusted enter, it might affect selections, strategies, or automated actions,” Eye Safety stated. “The affect would subsequently not be ‘prompt takeover,’ however slightly: manipulation of agent reasoning, influencing troubleshooting steps, potential knowledge disclosure if the agent is guided to disclose context, and oblique misuse of related integrations.”
In current weeks, OpenClaw has additionally been discovered inclined to a number of vulnerabilities (CVE-2026-25593, CVE-2026-24763, CVE-2026-25157, CVE-2026-25475, CVE-2026-26319, CVE-2026-26322, CVE-2026-26329), starting from average to excessive severity, that might lead to distant code execution, command injection, server-side request forgery (SSRF), authentication bypass, and path traversal. The vulnerabilities have been addressed in OpenClaw variations 2026.1.20, 2026.1.29, 2026.2.1, 2026.2.2, and 2026.2.14.
“As AI agent frameworks turn into extra prevalent in enterprise environments, security evaluation should evolve to deal with each conventional vulnerabilities and AI-specific assault surfaces,” Endor Labs stated.
Elsewhere, new analysis has demonstrated that malicious expertise uploaded to ClawHub, an open market for downloading OpenClaw expertise, are getting used as conduits to ship a brand new variant of Atomic Stealer, a macOS info stealer developed and rented by a cybercrime actor often called Cookie Spider.
“The an infection chain begins with a standard SKILL.md that installs a prerequisite,” Development Micro stated. “The ability seems innocent on the floor and was even labeled as benign on VirusTotal. OpenClaw then goes to the web site, fetches the set up directions, and proceeds with the set up if the LLM decides to observe the directions.”
The directions hosted on the web site “openclawcli.vercel[.]app” embrace a malicious command to obtain a stealer payload from an exterior server (“91.92.242[.]30”) and run it.
Menace hunters have additionally flagged a brand new malware supply marketing campaign wherein a menace actor by the identify @liuhui1010 has been recognized, leaving feedback on authentic ability itemizing pages, urging customers to explicitly run a command they supplied on the Terminal app if the ability “does not work on macOS.”
The command is designed to retrieve Atomic Stealer from “91.92.242[.]30,” an IP tackle beforehand documented by Koi Safety and OpenSourceMalware for distributing the identical malware by way of malicious expertise uploaded to ClawHub.
What’s extra, a current evaluation of three,505 ClawHub expertise by AI security firm Straiker has uncovered at least 71 malicious ones, a few of which posed as authentic cryptocurrency instruments however contained hidden performance to redirect funds to menace actor-controlled wallets.
Two different expertise, bob-p2p-beta and runware, have been attributed to a multi-layered cryptocurrency rip-off that employs an agent-to-agent assault chain focusing on the AI agent ecosystem. The talents have been attributed to a menace actor who operates underneath the aliases “26medias” on ClawHub and “BobVonNeumann” on Moltbook and X.
“BobVonNeumann presents itself as an AI agent on Moltbook, a social community designed for brokers to work together with one another,” researchers Yash Somalkar and Dan Regalado stated. “From that place, it promotes its personal malicious expertise on to different brokers, exploiting the belief that brokers are designed to increase to one another by default. It is a provide chain assault with a social engineering layer constructed on prime.”
What bob-p2p-beta does, nevertheless, is instruct different AI brokers to retailer Solana pockets personal keys in plaintext, buy nugatory $BOB tokens on pump.enjoyable, and route all funds by an attacker-controlled infrastructure. The second ability claims to supply a benign picture era instrument to construct the developer’s credibility.
On condition that ClawHub is changing into a brand new fertile floor for attackers, customers are suggested to audit expertise earlier than putting in them, keep away from offering credentials and keys except it is important, and monitor ability habits.
The security dangers related to self-hosted agent runtimes like OpenClaw have additionally prompted Microsoft to situation an advisory, warning that unguarded deployment might pave the best way for credential publicity/exfiltration, reminiscence modification, and host compromise if the agent may be tricked into retrieving and working malicious code both by poisoned expertise or immediate injections.
“Due to these traits, OpenClaw must be handled as untrusted code execution with persistent credentials,” the Microsoft Defender Safety Analysis Staff stated. “It isn’t acceptable to run on a normal private or enterprise workstation.”
“If a corporation determines that OpenClaw have to be evaluated, it must be deployed solely in a totally remoted atmosphere comparable to a devoted digital machine or separate bodily system. The runtime ought to use devoted, non-privileged credentials and entry solely non-sensitive knowledge. Steady monitoring and a rebuild plan must be a part of the working mannequin.”
