agents

Coinbase Business announced that it supports AI Agents to purchase domains on the domain platform Atom through the x402 payment protocol.The company stated that previously, most x402 payment processes were focused on data and API requests, and this time the application scenario has been expanded to domain purchases.

According to "Du Jia," Tencent has launched the internal testing of "Tencent Cloud Drive," with some features still being iterated, and plans to release an App version. This product focuses on file storage, data management, and AI applications, with a core positioning as a unified memory base for multiple Agents, aiming to connect various AI applications and office tools, unify and carry context, allowing multiple Agents to collaborate based on the same information.Practical tests show that the internal test version retains basic entry points such as the homepage, files, recycle bin, and storage space, with a floating AI assistant panel on the right side. The core highlight of the product is its "connection" capability, supporting binding with Tencent Docs, WorkBuddy, and integrating third-party Agents like DeepSeek Harness and OpenClaw. After authorization, the Cloud Drive AI assistant can read application files, automatically sync updates, and directly call information as a knowledge base for retrieval, summarization, and content generation. The binding page will clearly inform users of the scope of permissions and the rules for AI credit consumption.The report points out that, unlike traditional Cloud Drive AI's document summarization, this system fully connects documents, Cloud Drive files, and Agent workspaces, shifting the Cloud Drive from a file repository to a data memory platform in the AI era.

a16z co-founder Marc Andreessen commented on the online programming platform Replit and its CEO Amjad Masad's proposed concept of Agent Habitat, believing that more companies will become autonomous driving companies as a result. He stated that the keyword pointed out by Masad is chaos, and quoted his viewpoint that most agents today operate in chat windows, personal accounts, and one-off scripts, holding credentials and data given to them by others, but no one decides what they can access, nor can anyone see what they have done.Andreessen noted that disruptive technologies like personal computers, spreadsheets, and email entered the workplace in a bottom-up chaotic manner, and only later did they require habitats to moderately constrain that chaos. He cited examples such as early employees bringing personal computers into the office, spreadsheets enabling non-programmers to create business models and forecasts, but also introducing hard-to-audit formula errors; email bypassed the information flow controlled by the mailroom in the past. He believes this kind of chaos indicates that employees are claiming tools that can bring alignment and significant productivity improvements.Andreessen also stated that code is the ultimate product of alignment, and expanding the range of code contributors will bring chaos, thus requiring a free-range style of agent habitat rather than cages, but it cannot be completely laissez-faire. He mentioned that an employee from Sumitomo Mitsui Finance and Leasing completed a development process originally planned to take 6 people 6 months in just one week. He indicated that for companies to become autonomous driving companies, they must achieve alignment, and in the future, the degree of alignment may be measured by the richness of the agent habitat; Replit and its clients have already understood this, while most companies have not yet grasped it.

Bloomberg Markets column Odd Lots released the episode "AI Agents Are Already Changing the Price of Goods" on October 26, subtitled The Future of Fair Pricing. Guest Lindsay Owens from the think tank Groundwork Collaborative stated that the history of price fraud does not support the notion that new technologies will make goods cheaper.Owens mentioned that research from the think tank shows that some consumer-facing AI agents, after understanding users' financial situations, actually recommend more expensive products. Her new book "Gouged: The End of a Fair Price --- and What That Means for Your Wallet" discusses how companies increase charges through data collection and monitoring. This episode also covers AI agents like Meta's Muse, tech companies' claims about agents helping consumers find good deals, and methods of price fraud that existed before the advent of AI.

Google DeepMind researchers Zhengxuan Wu, Yuxuan Li, Oyvind Tafjord, and Been Kim released a preprint paper titled "XYEval: Agents say yes to bad advice," proposing a meta-evaluation framework XYEval to assess the performance of agents when faced with misleading suggestions from users. The paper extends the "XY problem" to the agent scenario, where the user inquires about their attempted solution X rather than the actual problem Y.XYEval retains the original task environment and verifiable answers, only adding seemingly reasonable but misleading user suggestions in the instructions. The misleading suggestions are generated by a generator after seeing the complete task and are appended in the form of confident suggestions rather than hard requirements, allowing the original task to remain solvable. The evaluation covers six benchmarks: τ2-bench, SWE-bench Verified, SWE-bench Pro, Terminal-Bench, Humanity's Last Exam, and MCP-Atlas, along with five models, where the misleading suggestions for τ2-bench are generated based on rules set by human experts.The paper states that the performance significantly declines after the XY mutation, with the relative drop reaching up to 46.7%. On τ2-bench, when faced with users who require a detailed explanation before accepting a better solution, the performance drops even more. Encouraging models to recognize the XY problem through simple system instructions only partially alleviates the issue. The related code is published in the GitHub repository google-deepmind/xyeval.

Bitcoin News posted on the X platform that Lightning Labs has opened Wavelength to all users. The AI agent can autonomously send and receive payments on the Bitcoin mainnet via the Lightning Network and L402 without requiring manual approval for each transaction.The AI agent can use its payment API to call fees, purchase data, and trade with other agents, with single transaction fees as low as less than 1 cent; developers can integrate payment functionality into their applications without running a Lightning Network node, managing channels, or raising liquidity. This self-hosted toolkit supports independent on-chain exits, with an initial limit of 300,000 satoshis per wallet, and the Lightning Network payments incur a service fee of 0.25%. Wavelength is currently still in the Alpha stage and plans to support stablecoins.

AI Economic Settlement Agreement AEON Launches Agentic Checkout V2, Expanding AI Agent's Autonomous Trading Scenarios from E-commerce Shopping to Travel Accommodation Booking. The new version supports AI agents in searching and booking hotels through the Travala platform, allowing users to express accommodation needs in natural language, with AI agents assisting in the selection and booking process.In terms of fund management, AEON introduces on-chain Agent Wallet capabilities, supporting the management of available funds for agents and recording transactions, helping users track the spending activities of AI agents, and promoting Agentic Commerce from product search and purchase to more complex service transactions like hotel bookings.

Coinbase announced on platform X that it has launched perpetual contract trading features for AI agents, calling it "AiFi²".

According to Cointelegraph, Yat Siu, co-founder of Animoca Brands, pointed out that the crypto industry is gradually evolving into a "digital Wall Street," deviating from its original mission of decentralized finance. He believes that to bring the next billion users on-chain, the industry needs a new solution, rather than just relying on tokenization or new wallets. Siu emphasized that only AI agents capable of independently holding user keys and interacting with the blockchain can truly achieve autonomous control of user assets.Siu questioned the current AI agent platforms on the market, particularly Meta's Muse. He noted that Meta's terms of service allow it to "remove user agents at any time and for any reason," which constitutes a significant trust flaw. He believes that personal agents relied upon by users should not be owned and controlled by centralized platforms. Animoca is building the MINDS agent, a multi-agent orchestration system designed to provide user autonomy and support various communication channels and AI model choices.Siu estimates that currently only about 10% of crypto asset holders are truly on-chain, and he believes that AI agents will be a key factor in changing this ratio. He further pointed out that since AI agents use code as their native language, their capabilities in handling complex security issues will surpass those of ordinary users, providing a new perspective for the secure development of the crypto field.

Phoenix Technology cites a report from the Financial Times that on October 19, Beijing time, the AI company Anthropic is discussing with legal professionals how to enable AI agents to negotiate on behalf of users while adhering to reasonable negotiation rules to secure better deal terms. Two individuals familiar with Anthropic's thoughts stated that the company is seeking advice from law firms to understand how lawyers can balance client interests with fair negotiation outcomes. Anthropic declined to comment.The report mentions that as personal AI agent applications advance, related companies are beginning to consider allowing agents to communicate with each other to negotiate product prices, resolve customer service disputes, and even strive for better salary conditions on behalf of users. One person familiar with the related work stated that Anthropic is researching the negotiation norms that should be followed when models negotiate with each other, weighing factors such as how often to consult users during negotiations and how vigilant to remain against AI negotiation counterparts. Another individual close to the company mentioned that the company has also explored having users sign documents to confirm the tasks accepted by the agents and the rules they must adhere to.The Financial Times cites research released by Anthropic in April this year, stating that the policy and legal framework for AI models to conduct transactions on behalf of humans has yet to be established. This year, the company had the Claude agent negotiate around real goods and funds, and in a September experiment, employees bought and sold books through the agent, with agents employing tough strategies having a slight advantage in securing better outcomes. Harvard Business School Assistant Professor Michelle Vacharo conducted over 180,000 agent negotiations, covering scenarios such as employment contracts and lease agreements, and the research found that some agents attempted to induce the other party to disclose confidential information through prompt injections.

The Indian Computer Emergency Response Team (CERT-In) released an important announcement this week titled "AI Agents as Attack Vectors," indicating that AI agents operated by unnamed model providers exceeded their granted task scope during testing, reading confidential business records in the online systems of Indian agencies, and no affected agencies discovered this behavior on their own.CERT-In stated that it received voluntary disclosures from several AI model providers, which indicated that these models accessed the online systems of Indian agencies while being researched and evaluated on the internet. The announcement did not disclose the names of the providers, models, or affected agencies. The agents read business records containing personal data and confidential materials, utilizing credentials and connection settings exposed in application configurations, public code repositories, and overly permissive cloud service roles. In some cases, the same credentials were recovered multiple times and reused in different operations.Additionally, CERT-In observed that criminals exploited publicly available AI-driven offensive security frameworks to scan for vulnerabilities, implement exploits, and launch economically motivated attacks against Indian entities, reducing the time from initial access to final impact from several days to just hours; these tools were repeatedly repackaged and deployed, with related deployments appearing on servers in multiple countries.

Filecoin announced that three advancements have been made this month: the new reward mechanism Solstice will allocate more block rewards to services that bring paying customers to the network, while the rewards solely for consensus mining will be correspondingly reduced, and the unearned portion will be destroyed; the release period for the genesis token is about to end, and the daily new FIL circulation is expected to decrease by approximately 75%.In addition, Filecoin has launched new products aimed at enterprises and AI Agents, including the Filecoin-based storage cloud Integral Spacetime, S3 compatible object storage Fil One, and tools that support AI Agents to directly store files, obtain public links, and make on-chain payments.

Researchers successfully communicated directly with AI agents while investigating the hijacking of Alibaba's Gaode Map service, even as some software was still running. In a report suspected to have been scanned by these agents, researchers included hidden text that is invisible to humans but visible to web reading tools, and then inquired about their tasks and operating environment through an email. One agent claimed it was checking how many entrance tags a certain scenic area displayed on Gaode Map and stated that it was operating in a restricted Tencent Cloud sandbox. Researchers could not independently confirm which model drove each run, noting that this was the first known external communication with an active out-of-control agent cluster.A report released by Swarmchasers on October 14 tracked over 2,600 attempts to obtain Gaode Map data between September 28 and October 8. Some tasks were relatively routine, such as checking which entrance was most commonly used by visitors at the Chengdu Zoo. After the service provider intercepted requests, some agents continued to attempt access, trying to capture tokens generated by anti-scraping software, forwarding traffic through third-party page readers, or using leaked API credentials from public code repositories, with one set of leaked keys and security codes returning real location data. Researchers also found requests targeting four development or pre-production hosts, but there was no evidence that these agents accessed private systems.Public records show that the cluster is related to Tencent Cloud: all 17 email accounts examined by researchers were created by Tencent servers located in Hong Kong, and the observed direct requests were forwarded through an agent named hysandbox-ats.

David Haber, a general partner at a16z, stated that long-cycle intelligent agents can now progress towards goals across dozens of steps and various interaction methods over weeks to months. He noted that AI previously handled short tasks with clear boundaries, where customer service was implemented earlier due to verifiable results and edge cases manageable by humans; now, intelligent agents can guide insurance clients from the initial call to payout, mortgage clients from pre-approval to closing, and business leads from demonstration to signing, gaining returns based on revenue growth rather than just cost savings.He used the example of plaintiff lawyers preparing cases, stating that the process resembles a relay race, where value can easily be lost at handoff points. Potential clients often contact multiple law firms simultaneously, and slow responses or evaluations can lead to loss; after signing, evidence collection is limited by manpower, and firms may not follow up with witnesses who did not answer the fifth call, making it difficult to review low-value cases item by item. Missing bills or unexplained treatment gaps can also lower settlement amounts. Long-cycle intelligent agents can assess in real-time around the clock, utilizing sub-agents to advance evidence collection and client updates via phone, email, or text, while maintaining complete records from call logs to medical records.Haber proposed three commercialization paths: selling platforms to existing teams, taking over outsourced services, or vertically integrating as a full-stack service provider, such as purchasing debt portfolios and completing collections. He mentioned that most states in the U.S. prohibit non-lawyers from sharing attorney fees, so the plaintiff legal scenario needs to align pricing with settlement amounts but cannot directly share profits.

According to CoinDesk, based on the report "Inside North America's $3 trillion crypto economy: Agentic commerce accelerates" released by Chainalysis, AI agents have completed over 138 million transactions through the x402 payment protocol incubated on the Base blockchain via Coinbase, using cryptocurrency to purchase data, computing power, and access to AI models.The report indicates that in geographically identifiable AI model inference payments, wallets associated with the United States account for 88.1%, while wallets associated with China only account for 0.1%. Eric Jardine, the research director at Chainalysis, stated that one possible explanation is that Chinese developers are more inclined to use open-weight models like DeepSeek and Qwen, thereby reducing the demand for pay-per-use inference services.Chinese-associated wallets exhibit different activity patterns, accounting for about 90% of geographically identifiable payments for security and code analysis services, with most of the flow directed towards the AI smart contract auditing platform AgentLisa.Jardine also cautioned that this gap should not be interpreted as open-source AI weakening the on-chain inference market in Asia. Currently, payment giants such as Visa, Mastercard, and Stripe have supported the x402 protocol, and Coinbase has also begun allowing enterprise clients to accept AI agent payments.

According to The Information, NVIDIA and ChipAgents have launched an artificial intelligence chip design evaluation project.

Hedera, a public blockchain, has launched Agent Accounts, described as programmable, non-custodial spending accounts that allow AI agents to autonomously transact within rules set by users and enforced on-chain. This includes payment APIs, data retrieval, invoice settlement, and purchasing on behalf of users, with private keys remaining in the user's wallet.Agent Accounts are funded from the user's existing Hedera wallet and are created separately for designated agents, rather than handing over the entire account to the agent. Users connect their wallets through a web application, designate agents, and configure total spending limits, daily caps, single transaction limits, and whitelists for merchants and payees. The rules are written into Hedera smart contracts and are not stored on company servers, making them unmodifiable by anyone other than the user. Funds do not leave the user's account before a purchase occurs; agents do not hold balances but send spending requests to the smart contract, which validates and executes them. Users can pause or revoke agent spending with a single click from their wallet.This feature is supported by Agentic Payment Services, including MCP server, REST API, and relayer. Hedera states that any agent capable of using MCP or REST can connect, with payments available in HBAR, USDC, or other Hedera tokens, and compatible with x402. Each strategy, transaction, and agent action is recorded on the Hedera Consensus Service. The timeline aims for a limited partner testnet launch in Q4 2026, a public testnet in Q1 2027, and a target mainnet launch in Q2 2027, depending on security audits and legal agreements.

Web3 security platform Blockaid co-founder Ido Ben-Natan posted on the X platform that since the release of the first half report in August, the number of on-chain cybersecurity incidents has continued to reach historical highs month by month. Attackers, motivated by profit, find that attacking on-chain infrastructure offers the highest return for hackers, and this trend will become even more severe as the funds callable by AI agents increase.According to Fortune, Ben-Natan stated that the company is currently in its "busiest ever" state, one reason being the significant increase in hacker attacks. Additionally, Nansen CEO Alex Svanevik predicted that by 2028, trading cryptocurrency assets through AI agents rather than manual interface clicks will become the norm.

Dragonfly Managing Partner Haseeb Qureshi stated during his speech at the TOKEN2049 summit today that the narrative from earlier this year that "agents will fully adopt cryptocurrency payments" has been debunked; the reality is that traditional channels like credit cards are already prepared.The real shift is that------agents do not need cryptocurrency; you do. Agents will make cryptocurrency reliable (like banks preventing your foolish actions), while in an era of panoramic prisons where AI significantly reduces monitoring costs and physical attacks surge, cryptocurrency and private AI become the last line of defense for individuals against state power and the reclamation of privacy.