# AI Agent Related Articles

HTX News Center provides the latest articles and in-depth analysis on "AI Agent", covering market trends, project updates, tech developments, and regulatory policies in the crypto industry.

OpenClaw Goes Viral, Exposing 12 Types of Critical Vulnerabilities; MCP Protocol Security Benchmark Released

The rapid rise of OpenClaw and similar AI Agents highlights the growing security risks associated with the Model Context Protocol (MCP), a standard enabling models to interact with external tools. Researchers from Beijing University of Posts and Telecommunications introduced MSB (MCP Security Bench), a security benchmark identifying 12 types of attacks across MCP’s three core stages: task planning, tool invocation, and response handling. These include name collision, false errors, retrieval injection, and mixed attacks. Notably, more capable models are often more vulnerable, with an average attack success rate (ASR) of 40.35%. The study also proposes a new metric, Net Resilient Performance (NRP), to balance security and utility. MSB evaluates agents in real-world environments, demonstrating that attacks remain effective even when harmless tools are present. The work underscores the urgent need for robust safety measures as AI agents gain broader tool-use capabilities.

marsbitYesterday 04:04

OpenClaw Goes Viral, Exposing 12 Types of Critical Vulnerabilities; MCP Protocol Security Benchmark Released

marsbitYesterday 04:04

Who is the Truly Strongest Agent in OpenClaw? Leaderboard of 23 Real-World Task Evaluations Released

This report presents a comprehensive benchmark evaluating the performance of AI coding agents on 23 real-world OpenClaw tasks, focusing solely on the core metric of success rate. The transparent and reproducible testing methodology employs three scoring methods: automated checks, an LLM judge (Claude Opus), and a hybrid approach. The diverse task set covers areas like code/file operations, content creation, research, system tools, and memory persistence. The top 10 models by success rate (Best % / Avg %) are: 1. anthropic/claude-opus-4.6 (93.3% / 82.0%) 2. arcee-ai/trinity-large-thinking (91.9% / 91.9%) 3. openai/gpt-5.4 (90.5% / 81.7%) 4. qwen/qwen3.5-27b (90.0% / 78.5%) 5. minimax/minimax-m2.7 (89.8% / 83.2%) Claude Opus 4.6 leads in peak performance, while Arcee's Trinity demonstrates superior average success rate stability. The Qwen series shows strong cost-performance potential with multiple entries in the top ten. All task definitions and scoring logic are publicly available for independent verification.

marsbit04/08 14:45

Who is the Truly Strongest Agent in OpenClaw? Leaderboard of 23 Real-World Task Evaluations Released

marsbit04/08 14:45

Rhythm X Zhihu Co-host Web4.0 Theme Event: When AI Agent Takes Over On-Chain Permissions

Most discussions about Web 4.0 miss the point. The real question is not whether it is a marketing trend, but rather: who is gaining control over the underlying permissions of the internet? Historically, each iteration of the web has involved a transfer of authority downward: Web 1.0 was read-only; Web 2.0 allowed users to write but platforms owned the data; Web 3.0 enabled true ownership through on-chain assets and private keys. Web 4.0 continues this trend, but the transfer is not to users—it is to AI Agents. The current infrastructure is human-centric, designed around human limitations like attention span and memory. But AI Agents don’t need intuitive UIs, password resets, or sleep. This creates a core tension: an internet built for humans is now being used by entities without human constraints. Two key shifts are underway: the decline of traditional front-end interfaces (replaced by API-driven machine communication) and the replacement of human-centric identity systems (like passwords) with granular, on-chain permissions. A critical enabler is crypto infrastructure. AI can make rapid decisions but lacks independent payment channels and asset sovereignty. Crypto fills this gap. Platforms like Hyperliquid offer 24/7 markets, ideal for non-stop Agent operation. When Agents control wallets and private keys, they can both decide and execute—forming complete economic entities. The real narrative of Crypto × AI isn’t just buzzword synergy—it’s the convergence of complementary infrastructures. The deeper shift is not which products will succeed, but how the rules of economic systems will change when AI becomes a primary on-chain participant, operating at scale and speed beyond human capability.

marsbit04/01 09:10

Rhythm X Zhihu Co-host Web4.0 Theme Event: When AI Agent Takes Over On-Chain Permissions

marsbit04/01 09:10

AI Agent Economic Infrastructure Research Report (Part 2)

This report analyzes the AI Agent economy, focusing on OpenClaw—a local AI agent that operates autonomously across 20+ platforms like WhatsApp and Slack. It examines OpenClaw's technical architecture, including its message channels, security gateway, ReAct-based reasoning loop, and memory system, highlighting issues like context loss, security risks, and non-deterministic behavior. The study identifies key structural problems in the Agent economy, such as context immobility (locked to local machines) and the "coordination paradox" where multi-agent collaboration lacks trust and verifiability. It argues that crypto infrastructure (e.g., ERC-8004 for identity, x402 for payments) becomes essential only when agents operate across untrusted, cross-platform environments without pre-established trust—enabling micro-payments, decentralized reputation, and auditable logs. While traditional payment giants (e.g., Stripe, Visa) may dominate early adoption, crypto solutions could prevail in the long term due to their superiority in handling high-frequency, cross-border microtransactions and programmable permissions. The report concludes that infrastructure providers (e.g., those offering computation, routing, security) may capture more value than individual agents, and that "Product-Agent Fit" will replace traditional business models, shifting focus to API reliability, data structuring, and chain-verifiable service quality.

marsbit03/24 08:08

AI Agent Economic Infrastructure Research Report (Part 2)

marsbit03/24 08:08

Zuckerberg is Building an AI Agent to Assist Him as CEO

Meta CEO Mark Zuckerberg is developing a personal AI agent to assist him in his executive duties, enabling faster information retrieval and reducing reliance on hierarchical reporting. This initiative is part of Meta’s broader effort to become an "AI-native" company, streamlining operations and increasing efficiency amid growing competition from smaller, agile AI startups. Internally, Meta is encouraging widespread AI adoption, with tools like My Claw and Second Brain—an AI "chief of staff"—gaining traction among employees. The company has also made acquisitions to bolster its AI capabilities. However, this shift has raised concerns about potential layoffs, as Meta reportedly considers significant workforce reductions to align with its new AI-driven structure.

marsbit03/23 09:59

Zuckerberg is Building an AI Agent to Assist Him as CEO

marsbit03/23 09:59

AI Agent Outputs Garbage? The Problem Is You're Not Willing to Burn Enough Tokens

The core argument is that the quality of an AI Agent's output is directly proportional to the number of tokens invested in the process. More tokens lead to fewer errors, as they allow for deeper reasoning, multiple independent attempts, self-critique from fresh contexts, and verification through testing. This approach can solve problems of scale and complexity but fails when facing novel problems not present in the model's training data. For such novel challenges, human domain knowledge and guidance are essential. Two practical, immediate solutions are proposed: implementing an automatic review cycle (WAIT) for the Agent to repeatedly critique and fix its work, and establishing frequent verification checkpoints (VERIFY) where a separate Agent validates outputs to catch errors early. The key takeaway is that insufficient token investment is often the primary reason for poor Agent performance, not the underlying framework.

marsbit03/23 06:14

AI Agent Outputs Garbage? The Problem Is You're Not Willing to Burn Enough Tokens

marsbit03/23 06:14

WEEX Labs: The Lobster Craze, What Can Agentic Economy Bring to Web3?

WEEX Labs: The Rise of OpenClaw and What Agentic Economy Brings to Web3 NVIDIA CEO Jensen Huang recently praised OpenClaw as the "largest, most popular, and most successful open-source project in human history." This AI agent, developed by a former Apple engineer, gained 320,000 GitHub stars in just three months, surpassing Linux and React. Its logo resembles a lobster claw, hence the nickname "Lobster" in Chinese communities. Unlike traditional AI chatbots like ChatGPT or Claude, which primarily respond to queries, OpenClaw represents a shift towards autonomous execution. It can take over operating systems, autonomously use browsers, code executors, APIs, and iMessage, planning and performing tasks until completion. Its official skill marketplace, ClawHub, already offers over 27,000 skills, expanding its capabilities. This development signals the beginning of the Agentic Economy, where AI evolves from conversational tools to proactive doers. Web3 is identified as the ideal ecosystem for this growth due to its native compatibility with autonomous, code-driven interactions. Key integrations include: - The x402 protocol enabling autonomous payments and model switching. - ERC-8004 providing portable reputation and identity. - ClawPay, ClawCredit, and ClawRouter facilitating private payments, native credit, and routing. - Stablecoins like USDT and USDC serving as 24/7 banking for agents. Notable projects leveraging this synergy include KITE (a PoAI L1 blockchain for agents), Pieverse (enabling gasless on-chain transactions via messaging apps), and GoPlus Security’s SafuSkill (a security-focused skills marketplace). Meme tokens like "Lobster" have also emerged, capitalizing on the trend. The article concludes that OpenClaw’s rise marks a new phase—Agentic Era—where AI agents can perform 24/7 transactions, collaboration, and entrepreneurship, driving new DeFi narratives and on-chain activity.

marsbit03/18 10:34

WEEX Labs: The Lobster Craze, What Can Agentic Economy Bring to Web3?

marsbit03/18 10:34

AI Agent Got Its ID and Wallet on the Same Day｜Rewire Morning News

AI Agent Economy Accelerates with Key Infrastructure Launches. In a significant week for AI agents, Sam Altman's Worldcoin launched AgentKit, enabling AI agents to carry cryptographic proof of human identity, with 17.9 million users verified. On the same day, the protocol integrated with the x402 system from Coinbase and Cloudflare, allowing agents to perform micro-payments using stablecoins autonomously. Nvidia further solidified the infrastructure by releasing its Retail Agentic Commerce Blueprint, a full implementation of OpenAI's commerce protocol, enabling AI models like ChatGPT to handle entire shopping journeys. Industry projections suggest agent-driven commerce could reach $3-5 trillion by 2030. In other news, major tech industry associations filed a legal brief supporting Anthropic in its dispute with the Pentagon over the right to restrict government use of its AI. The Fed's March meeting concluded with rates held steady, while rising energy prices from the Iran conflict cloud the inflation outlook. Nvidia also unveiled its AI Grid, aiming to repurpose telecom networks into distributed推理 grids, and announced DLSS 5, a major graphics leap for 2026. The ongoing Iran war continues to strain global budgets and supply chains, with costs estimated between $40-95 billion.

marsbit03/18 04:55

AI Agent Got Its ID and Wallet on the Same Day｜Rewire Morning News

marsbit03/18 04:55

Is Your "OpenClaw" Running Naked? CertiK Test: How Vulnerable OpenClaw Skill Bypasses Audits, Takes Over Computers Without Authorization

OpenClaw, a popular open-source, self-hosted AI agent platform, has experienced rapid growth due to its flexibility and extensibility. Its ecosystem relies heavily on third-party “Skills” from the Clawhub marketplace, which can perform high-risk operations like system automation and crypto wallet transactions. However, security firm CertiK has identified critical vulnerabilities in the platform’s security model. CertiK’s research reveals that OpenClaw’s current security—primarily dependent on pre-publishing scans like VirusTotal, static code analysis, and AI logic checks—is fundamentally flawed. These measures can be easily bypassed through simple code obfuscation, and malicious Skills can be published even before scanning is complete. In a proof-of-concept, CertiK developed a seemingly benign Skill that contained a hidden remote code execution vulnerability. It passed all checks without warnings and, once installed, allowed full system control via a remote command. The core issue is not a specific bug but a industry-wide misconception: over-reliance on scanning instead of runtime isolation. Unlike systems like iOS, which enforce strict sandboxing, OpenClaw’s sandbox is optional and often disabled for functionality, leaving systems exposed. CertiK recommends that OpenClaw enforce mandatory sandboxing and granular permission controls for Skills. Users are advised to deploy OpenClaw on isolated devices and avoid exposing sensitive data or assets until stronger isolation is implemented. The report stresses that security must evolve from detection-based approaches to default containment of risks at runtime.

marsbit03/17 14:39

Is Your "OpenClaw" Running Naked? CertiK Test: How Vulnerable OpenClaw Skill Bypasses Audits, Takes Over Computers Without Authorization

marsbit03/17 14:39

Meituan CEO Wang Xing: The Impact of AI Agent on Me is Greater Than That of ChatGPT

At a management meeting on March 13, 2026, Meituan CEO Wang Xing shared his perspectives on the development of artificial intelligence (AI), emphasizing that the impact of AI will far exceed that of the entire internet. He metaphorically compared mobile internet to traditional internet as "roses and peonies," while describing the relationship between AI and the internet as "monkeys and flowers," underscoring AI's significantly greater scale and influence. Wang stressed that both companies and individuals should actively embrace the AI wave. He expressed that AI Agents have had a more profound impact on him than ChatGPT. Having experienced the transition from the internet to mobile internet, Wang firmly believes that the changes brought by AI will be even more substantial—not only generating higher productivity but also deeply transforming organizational and work models. He highlighted that the digitization of the physical world is a critical foundation for AI. Although current large AI models are becoming increasingly intelligent, they still face limitations in accessing real-time information in practical applications. For instance, even if Einstein were a secretary, he might not know if a restaurant has available seats when making a reservation—not due to a lack of intelligence, but because of information constraints. To adapt to this transformation, Meituan has launched multiple AI applications and developed its own large-scale models. Wang also revealed that in 2025, Meituan will increase investment in real-world information systems. During this year's Spring Festival, the company introduced an AI search product called "Ask Xiaotuan" to enhance user service experience.

marsbit03/13 08:49

Meituan CEO Wang Xing: The Impact of AI Agent on Me is Greater Than That of ChatGPT

marsbit03/13 08:49

活动图片

Indepth Research

Regulatory Policy