IronClaw emerged as a Rust-based alternative to address OpenClaw's security vulnerabilities.

Near.AI co-founder Illia Polosukhin announced IronClaw, a new system built in the Rust programming language to address security vulnerabilities in the AI assistant OpenClaw. OpenClaw had drawn attention as a powerful interface capable of controlling users’ browsers, devices, and crypto wallets; however, these capabilities had faced serious criticism, particularly due to the risk of private key exposure.

Security as a Core Design Principle

IronClaw ensures that each component runs in its own isolated WebAssembly environment. This architecture prevents a compromised component from damaging the rest of the system. Additionally, the language model is not granted direct access to user credentials. Crypto keys and other sensitive data are stored in an encrypted secure vault and are only unlocked for limited, authorized use on specific websites.

Polosukhin stated that he designed the system based on the principle of “never giving the LLM access to secrets.” This approach aims to prevent the financial losses users experienced while using OpenClaw from recurring.

Rapid Development and Real-World Impact

Near.AI CEO George Xian Zeng recounted that Polosukhin completed the core architecture of IronClaw in a single evening, all while feeding his baby. The developer, who made 74 GitHub commits last week, plans to release the system in beta on the Near.AI platform within a few weeks.

Currently, Near.AI Cloud serves those requesting access to run OpenClaw within a secure Trusted Execution Environment. In this environment, user data remains encrypted, and even Near.AI cannot access it.

Online Skills Marketplace: Security Is Critical

In ClawHub, the skills marketplace within the OpenClaw ecosystem, 341 existing skills were found to contain malicious code. These skills are used to harvest users’ passwords and personal data. Zeng emphasized the current challenge of securing the marketplace, stating, “Anyone can create a skill—that’s both an advantage and a risk.” A corporate curation system is planned for the near future to address this issue.

Meanwhile: Olas Unleashes AI Agents on Polymarket

Meanwhile, the company Olas has deployed AI systems designed for prediction markets onto Polymarket. These bots, named Polystrat and tested on the Omen platform, achieve higher success rates than predicted by using news sources, data streams, and other tools for predictions that resolve within four days. According to the data, these systems achieve success rates of 59–64% in scientific and commercial predictions, but drop to 38–49% for sports, fashion, and social events.

The core architecture of Polystrat ensures security by completely excluding the ability to control user wallets. The bots are not authorized to move user funds under any circumstances. This design preserves only prediction-focused functionality, not assistant-like capabilities.

Broader Technology Landscape

The first months of 2026 in the AI space are defined by security-focused innovations. Amazon’s “Search Party” feature, based on Ring cameras, has reignited neighborhood surveillance concerns. In response, Wyze released a comedy video mocking Amazon. During the same period, 16 tech companies highlighted AI products in Super Bowl commercials. Some analysts note that such advertising surges have historically preceded short-term downturns in technology sectors.

AI has now firmly established itself not only as digital assistants but also in media production. According to a McKinsey report, tools that automatically convert scripts into shooting plans are creating speed and cost advantages in film and TV production. This transformation is forcing production companies to adopt AI technologies out of necessity.