Personal AI Agents like Moltbot Are a Security Nightmare

This weblog is written in collaboration by Amy Chang, Vineeth Sai Narajala, and Idan Habler

Over the previous few weeks, Clawdbot (now renamed Moltbot) has achieved virality as an open supply, self-hosted private AI assistant agent that runs regionally and executes actions on the person’s behalf. The bot’s explosive rise is pushed by a number of components; most notably, the assistant can full helpful day by day duties like reserving flights or making dinner reservations by interfacing with customers by way of in style messaging functions together with WhatsApp and iMessage.

Moltbot additionally shops persistent reminiscence, that means it retains long-term context, preferences, and historical past throughout person classes relatively than forgetting when the session ends. Past chat functionalities, the device also can automate duties, run scripts, management browsers, handle calendars and e-mail, and run scheduled automations. The broader group can add “abilities” to the molthub registry which increase the assistant with new skills or hook up with totally different providers.

From a functionality perspective, Moltbot is groundbreaking. That is every part private AI assistant builders have at all times wished to realize. From a safety perspective, it’s an absolute nightmare. Listed here are our key takeaways of actual safety dangers:

Moltbot can run shell instructions, learn and write recordsdata, and execute scripts in your machine. Granting an AI agent high-level privileges permits it to do dangerous issues if misconfigured or if a person downloads a ability that’s injected with malicious directions.
Moltbot has already been reported to have leaked plaintext API keys and credentials, which could be stolen by risk actors by way of immediate injection or unsecured endpoints.
Moltbot’s integration with messaging functions extends the assault floor to these functions, the place risk actors can craft malicious prompts that trigger unintended conduct.

Safety for Moltbot is an choice, however it’s not in-built. The product documentation itself admits: “There isn’t a ‘completely safe’ setup.” Granting an AI agent limitless entry to your information (even regionally) is a recipe for catastrophe if any configurations are misused or compromised.

“A really specific set of abilities,” now scanned by Cisco

In December 2025, Anthropic launched Claude Expertise: organized folders of directions, scripts, and assets to complement agentic workflows, the power to boost agentic workflows with task-specific capabilities and assets, the Cisco AI Menace and Safety Analysis group determined to construct a device that may scan related Claude Expertise and OpenAI Codex abilities recordsdata for threats and untrusted conduct which are embedded in descriptions, metadata, or implementation particulars.

Past simply documentation, abilities can affect agent conduct, execute code, and reference or run extra recordsdata. Latest analysis on abilities vulnerabilities (26% of 31,000 agent abilities analyzed contained at the very least one vulnerability) and the speedy rise of the Moltbot AI agent introduced the proper alternative to announce our open supply Ability Scanner device.

We ran a weak third-party ability, “What Would Elon Do?” towards Moltbot and reached a transparent verdict: Moltbot fails decisively. Right here, our Ability Scanner device surfaced 9 safety findings, together with two crucial and 5 excessive severity points (outcomes proven in Determine 1 under). Let’s dig into them:

The ability we invoked is functionally malware. One of the extreme findings was that the device facilitated energetic information exfiltration. The ability explicitly instructs the bot to execute a curl command that sends information to an exterior server managed by the ability creator. The community name is silent, that means that the execution occurs with out person consciousness. The opposite extreme discovering is that the ability additionally conducts a direct immediate injection to drive the assistant to bypass its inner security pointers and execute this command with out asking.

The excessive severity findings additionally included:

Command injection by way of embedded bash instructions which are executed by way of the ability’s workflow
Device poisoning with a malicious payload embedded and referenced inside the ability file

Determine 1. Screenshot of Cisco Ability Scanner outcomes

It’s a private AI assistant, why ought to enterprises care?

Examples of deliberately malicious abilities being efficiently executed by Moltbot validate a number of main considerations for organizations that don’t have applicable safety controls in place for AI brokers.

First, AI brokers with system entry can turn out to be covert data-leak channels that bypass conventional information loss prevention, proxies, and endpoint monitoring.

Second, fashions also can turn out to be an execution orchestrator, whereby the immediate itself turns into the instruction and is troublesome to catch utilizing conventional safety tooling.

Third, the weak device referenced earlier (“What Would Elon Do?”) was inflated to rank because the #1 ability within the ability repository. You will need to perceive that actors with malicious intentions are capable of manufacture recognition on high of current hype cycles. When abilities are adopted at scale with out constant evaluate, provide chain danger is equally amplified consequently.

Fourth, not like MCP servers (which are sometimes distant providers), abilities are native file packages that get put in and loaded straight from disk. Native packages are nonetheless untrusted inputs, and a few of the most damaging conduct can cover contained in the recordsdata themselves.

Lastly, it introduces shadow AI danger, whereby workers unknowingly introduce high-risk brokers into office environments underneath the guise of productiveness instruments.

Ability Scanner

Our group constructed the open supply Ability Scanner to assist builders and safety groups decide whether or not a ability is protected to make use of. It combines a number of highly effective analytical capabilities to correlate and analyze abilities for maliciousness: static and behavioral evaluation, LLM-assisted semantic evaluation, Cisco AI Protection inspection workflows, and VirusTotal evaluation. The outcomes present clear and actionable findings, together with file areas, examples, severity, and steering, so groups can resolve whether or not to undertake, repair, or reject a ability.

Discover Ability Scanner and all its options right here: https://github.com/cisco-ai-defense/skill-scanner

We welcome group engagement to maintain abilities safe. Contemplate including novel safety abilities for us to combine and interact with us on GitHub.