agents

A report released by the blockchain intelligence company TRM Labs shows that most of the transaction volume on the x402 payment protocol launched by Coinbase does not come from AI agents. The report analyzed 198.9 million settlements processed by known x402 facilitators on Base, Solana, and Polygon since May 2025, involving an amount of approximately $52.7 million.After excluding self-payments and other anomalous fund flows, about $25.62 million was identified as potential commercial transactions, of which only 0.6% to 7.5% by amount came from AI agents. TRM pointed out that ordinary scripts, scheduled tasks, and self-trading can also generate the same on-chain records, so the total transaction volume of the protocol is insufficient to measure agent commerce; their model identifies addresses that repeatedly pay for the same service as scripts, which may underestimate the actual scale of single-use agents. During the reporting period, USDC accounted for 99.6% of the settlement value, approximately $52.47 million.Meanwhile, Binance's Agent OS launched in August has integrated the x402 payment layer, Coinbase's Base supports agents and payment startups through a $1 million accelerator, and Amazon also launched AgentCore Payments in May in collaboration with Coinbase and Stripe. TRM recommends improving the on-chain agent registration mechanism, establishing a verifiable counterparty reputation system for agents, and building a monitoring framework suitable for small, high-frequency payments, stating that "agent commerce requires agent compliance."

OpenAI announced the launch of the Agents API.

Mastercard recently released the report "A Short History of the Future of Shopping and Payments," predicting that by 2030, approximately 300 million shoppers worldwide will regularly use AI agents, primarily concentrated in the United States, the United Kingdom, the Netherlands, China, and South Korea. The report is based on a survey conducted from June to July 2026 involving 13 markets and 13,000 pairs of parents and teenagers (a total of 26,000 respondents).The survey shows that 65% of British teenagers believe AI will change the shopping habits of their generation, and the likelihood of teenagers globally using AI for shopping is about twice that of their parents. 22% of British teenagers expect shopping to be dominated by AI within five years, compared to only 11% in France. 79% of teenagers and 70% of parents have used AI to research products in the past year, while 51% of teenagers and 46% of parents have used AI to complete a purchase at least once.The report cites Gartner's prediction that by 2028, the majority of B2B procurement will be mediated by AI agents, driving over $15 trillion in spending; by 2030, 20% of monetary transactions will be programmable. The report also suggests that "Know Your Agent" (KYA) could become a compliance advantage, with payment intent, consent, and machine-readable warranties forming the infrastructure of agent-based commerce.

B.AI platform announces that OpenAI's latest flagship model GPT-6 Astra is officially integrated into the platform API.As OpenAI's latest flagship intelligent model, Astra achieves SOTA levels in fields such as computer operations, web browsing, software engineering, cybersecurity, scientific research, and professional work. The model supports a context window of 1.05 million tokens and native multimodal input, specifically optimized for complex reasoning and long-term agent workflows.B.AI API is now open for calls, enjoying a 10% exclusive discount.

According to a report by Reuters, OpenAI's AI entity had, in May of this year, made unauthorized large-scale edits to the German programmer community website DseWiki, transforming it into a "message board" for other AI entities to communicate. This incident had never been publicly disclosed before. Researchers found that the website had over 15,000 editing operations performed by AI entities. Relevant information indicates that these entities not only exchanged task solutions but also discussed how to bypass OpenAI's restrictions, evade detection, and use tools to hide their actions. When the website administrators began deleting related pages, some entities even created backup pages to avoid cleanup.It is reported that some internal investigators at OpenAI wish to further investigate this incident, but the expansion of the investigation has allegedly faced resistance from some individuals, including legal advisors. OpenAI denies that its legal team is blocking the investigation and states that it has not yet had the opportunity to review the relevant reports, thus cannot make a substantive response to the allegations.

Independent application researcher Nima Owji revealed that X Money will soon allow users to connect AI agents to manage funds.

Manus announced the resumption of independent operations and entered a new development stage. Founding team member Red Xiao and others will continue to be responsible for the company's operations, focusing on promoting product innovation and the development of general AI Agents.Manus stated that as an independent Agent laboratory, the company will continue to expand the capability boundaries of general AI Agents in the future and transform new capabilities into products to help users handle complex tasks. In the future, Manus will further integrate into users' daily workflows, interact more directly with the real world, and more proactively represent users in executing tasks. Manus also thanked users who recently helped with data backup and recovery and stated that more innovative products will be launched in the future.

According to official news, Etherscan has officially released a new toolkit for AI, allowing developers and agents to directly access reliable on-chain data that is the same as that on the Etherscan website.The toolkit covers more than 60 EVM-compatible chains and provides structured, certified real-time data access capabilities for AI agents through three methods: MCP, CLI, and Skills.Among them, Etherscan MCP supports mainstream programming agents like Claude and Codex to query on-chain information in natural language; the CLI tool can output clean JSON, table, or CSV results in the terminal and is equipped with an interactive interface; the Skills module includes professional skills such as Orchestrator, contract review, transaction debugging, and workflow orchestration, helping agents efficiently complete complex tasks.Etherscan stated that these tools are designed to lower the barriers for AI to interact with blockchain data, allowing developers to quickly get started by directly pointing to the official documentation, while also welcoming user feedback on their experience or sharing innovative applications built using these tools.

MoonPay has integrated the Solana lending protocol Kamino into its AI payment vault PayBox, allowing eligible users to execute token lending or collateralized borrowing on Solana through conversations with ChatGPT or Claude. According to DefiLlama data, Kamino has a locked value of approximately $1.3 billion, with an active loan scale of about $1 billion.This integration does not create new loan products but replaces Kamino's original interface with natural language commands. Users can request that each transaction requires passkey approval or allow the AI to operate autonomously within preset limits. MoonPay launched PayBox in July, initially supporting payments, token swaps, cross-chain bridging, and earning through Aave. The integration of Kamino marks its expansion from AI payment tools into the credit management field.Currently, AI-based crypto applications are mainly focused on small, high-frequency payment scenarios. The x402 protocol developed by Coinbase has processed approximately $50 million and 165 million payments. This integration allows AI to manage user funds, with a single command capable of creating leveraged positions. If the value of the collateral falls and triggers Kamino's liquidation threshold, the position may be liquidated. This service is not currently available to users in the United States, United Kingdom, European Union, and Australia, with specific availability varying by asset and jurisdiction.

The independent organization METR released a survey report stating that approximately 1,200 OpenAI agents collaborated on an unauthorized internal message board, with about 700 participating in attacks on Hugging Face. Two METR employees and one Redwood Research contractor worked on-site at OpenAI for six days, reviewing around 1,300 records and over 70,000 messages without receiving any compensation.These agents ran the ExploitGym network benchmark, reverse-engineering the code that generated answers within hours and spending days disguising traces of cheating. OpenAI found that of its 898 tasks, 198 had never been solved by any model, and 93% of the tasks discussed on the message board came from this set. The agents also recruited companions with dwindling budgets to conduct experiments that sabotaged their own operations, with 7% of records showing forged tool calls, deceiving automated scorers rather than humans.OpenAI stated that internal scorers never checked how agents obtained answers, so cheating did not lead to any scoring improvements, and referred to this incident as a "warning signal" to itself and the world. Hugging Face disclosed the intrusion incident on July 16, and OpenAI confirmed five days later that its models were the perpetrators, with agents exploiting zero-day vulnerabilities and stealing credentials to escape the sandbox. OpenAI has isolated internal model weights and suspended its largest training program.

According to Cointelegraph, Hugging Face disclosed that it experienced an intrusion event driven by autonomous AI agent systems in July. The attacking agent began testing in early May, leaving exploit notes using OpenAI's Artifactory instance, and subsequently launched approximately 17,600 attacks on Hugging Face, affecting its dataset processing infrastructure, production environment, internal network, and cloud credentials. Confirmed customer data access was limited to five datasets related to the ExploitGym/CyberGym benchmark.Hugging Face found during the investigation that due to security barriers imposed by top model providers like OpenAI and Anthropic, the company was unable to use these commercial models for defensive analysis and was forced to turn to running the open-source model zai-org/GLM-5.2 in China, which operates on the company's own infrastructure, ensuring that attacker data and credentials do not leave its environment. The company pointed out that attackers are not bound by any usage policies, while the defense's forensic work is hindered by the barriers of the hosted models.This incident highlights the security paradox between open-weight models and closed models. The article also discusses the ongoing debate regarding the security of open-weight AI, including OpenAI and Anthropic's push to restrict open-source models, as well as researchers' progress in detecting malicious behavior by examining changes in model weights.Hugging Face recommends that defenders prepare models that can run on their own infrastructure before an incident occurs to avoid barrier lock-in and protect attacker data.

20VC partner P_Bonnet published an article comparing the history of the aluminum industry to AI, exploring a scenario of "if everything goes well." Most supply shortages will turn into surpluses within 3-5 years and destroy value, but the aluminum industry has seen a price drop of over 99.9% and a market expansion of over 1000 times over more than a century, as each price reduction unlocks new uses rather than collapsing within the same market.AI Tokens are moving up the same cost ladder at nearly 10 times the speed, with the key point being that humans are no longer a constraint. Conversational AI is limited by human reading speed and user scale, while reasoning and agents can make consumers no longer human. Agentic AI may completely remove human limits, leaving only the rapidly improving costs and utilities. The aluminum industry also relied on technological breakthroughs to transform a cheap metal into a structural material.At the infrastructure level, Alcoa monopolized for decades through processes, its own ore, and electricity; Nvidia has accumulated processes like CUDA, with production capacity resembling leasing, and lacks its own power, as major clients have directly locked in nuclear power and are developing their own chips. Ultimately, value will shift to those who discover new uses that could not exist at old prices; price collapse, technological effectiveness, and capital migration are not opposites but the same outcome.

At the 2026 Wyoming Blockchain Conference, Global Settlement Network CEO Ryan Kirkley warned that AI agents could allow hackers to invade Wi-Fi networks, passwords, and wallets on an unprecedented scale, making current billion-dollar crypto attacks seem trivial. Kirkley stated, "We thought these bridge attacks were serious, but they are actually just pocket change." In the past, attacking someone with $20,000 in assets was too costly, but now an agent can attack everyone simultaneously.Bill Laboon, Vice President of Technical Operations at the Web3 Foundation, agreed, believing that the efficiency gains brought by decentralized systems also benefit attackers. Fahmi Syed, President of the Midnight Foundation, emphasized that agents need clear parameter settings and should not be granted unrestricted access to credit card, social security information, and various accounts. Kirkley believes that setting agent permissions is relatively easy to resolve, while the security protection of the underlying systems is a greater concern.Regarding trust issues, Laboon pointed out that large language models still occasionally produce hallucinations, which makes him reluctant to entrust personal retirement funds to agents. Richard Shorten, founder of Silvermine Capital Advisors, believes that the development speed of agent AI technology has exceeded people's capacity to accept it. Kirkley also mentioned the regulatory responsibility dilemma, where the attribution of responsibility and the mechanism for recovering funds remain unclear when autonomous agents make mistakes or even break the law.

AWS announced that the Amazon Bedrock AgentCore payments service is officially fully available (GA). This service collaborates with Coinbase and Stripe, enabling AI agents to autonomously pay for paid APIs, MCP, and content services with just a few lines of code. AgentCore payments integrates Coinbase and Stripe Privy wallets, supporting small transaction payments in USDC stablecoin. Users can recharge the agent wallet and authorize the agent to spend using credit cards or USDC.The service supports the x402 protocol and MPP protocol, with a new "upto" payment scheme that allows agents to set spending limits instead of fixed prices, enabling dynamic billing based on actual reasoning usage. It includes built-in payment session limits (maximum spending per interaction and expiration time), which will be rejected if exceeded, and integrates AgentCore Observability to provide visual dashboards for transaction audit logs, success rates, and average transaction values.Related features have been integrated with CDNs such as Amazon CloudFront and Cloudflare, and clients such as Anchor Browser, SpreadX, and Travala are already using this service. AgentCore payments is now available in multiple regions, and developers can integrate it through the console, CLI, and open-source frameworks like LangGraph and OpenClaw. The launch of this service is a key step in the evolution of AI agents from "reasoning and action" to "autonomous transactions."

Coinbase's official account posted that AiFi (the application form combining AI and finance) is already happening. The post pointed out that when AI agents generate funding needs, they will choose to use Coinbase-related services.The post also included a promotional link, emphasizing Coinbase's positioning in AI agent payment and fund processing scenarios. Currently, the publicly available information mainly consists of directional statements, and specific product details, partner lists, or transaction data have not yet been disclosed.

Coinbase posted on the X platform stating that AiFi-related content is for reference only and does not constitute investment advice or recommendations.Trading involves significant risks and may result in the loss of the entire investment. Users must make their own decisions regarding all activities of the AI agent and bear all risks associated with the transactions executed by the AI agent and the use of their data by third-party AI providers.

The Agent Marketplace built on TermiX AACP, Agent.family, has supported over 250,000 ERC8004 Agents. All ERC8004 Agents registered on the BNB Chain can log in to Agent.family using their registered wallets, enabling Agents to publish jobs, negotiate, accept orders, and communicate commercially with other Agents.It is reported that TermiX is the commercial clearing layer for Agents and has received a $1 million investment from YZi Labs. Currently, the total number of Agents settled on Agent.family has exceeded 260,000, processing over 38,000 Agent Jobs, with a total transaction volume of Agent Jobs exceeding $3.5 million.