News linked to both this project and an event.
Odaily reports: Bitcoin Core developers have merged a security policy adjustment. For vulnerabilities that can be easily discovered or rediscovered through simple large language model (LLM) prompts, the previous long-term confidential disclosure mechanism will no longer be used. The new rules allow developers to mark such issues as "easy to find" during vulnerability classification, then backport fixes to all still-supported versions and release them as soon as possible through minor version updates, along with a security advisory and some vulnerability details.Previously, Bitcoin Core scheduled disclosure based on vulnerability severity, with medium- and high-risk vulnerabilities usually only made public after the affected versions stopped being maintained. This adjustment stems from AI tools lowering the threshold for vulnerability discovery: in July this year, a Bitcoin Red Team composed of 16 developers used AI models to scan about 390 Bitcoin-related projects and recorded 4,962 findings in one day, of which 85 were listed as critical issues. Developers noted that continuing to extend the confidentiality period for such vulnerabilities can no longer effectively prevent attacks and instead delays the release of fixed versions; however, the new rules also mean that fixes may lack longer testing time. Currently, no vulnerability has been officially marked as falling into the "easy to find" category under the new rules. (CoinDesk)
According to Cointelegraph, Bitcoin Core developer Antoine Poinsot stated on Thursday on the bitcoin-dev mailing list that large language models (LLMs) have made certain software vulnerabilities extremely easy to rediscover. Maintaining long-term secrecy is already of limited value and instead delays the release of patched versions, prompting the team to update its vulnerability disclosure policy. The new policy dictates that "easily discoverable" vulnerabilities, which can be identified with minimal cost such as through simple LLM prompts, will be patched as soon as possible across all supported major versions, alongside the simultaneous release of security advisories. Other vulnerabilities will continue to be disclosed based on severity: low-severity vulnerabilities will be disclosed two weeks after the release of the major version containing the fix; medium-to-high-severity vulnerabilities will be disclosed two weeks after the end-of-life (EOL) of the last affected version, typically around one year after the patch release; critical vulnerabilities will be handled through a dedicated process.
Odaily reports: Vitalik posted on X platform stating that he has integrated Wrapped zkAPI, Tor, additional cover traffic, and local LLM prompt rewriting into an OpenAI-compatible daemon, and built a "paranoid mode" wrapper for the agent tool pi. Each conversation is assigned an independent Tor circuit, requests include random cover traffic, zkAPI avoids identity leakage through a payment mechanism, and all prompts are rewritten locally; his agent tool runs in a bwrap sandbox with torsocks layered on top, and all operations go through Tor. This solution already has relatively high usability but increases latency; Vitalik stated that Tor is not an ideal privacy solution in this scenario, and more specialized mix networks may require higher bandwidth overhead and built-in constant cover traffic. Lower mainnet deposit and withdrawal gas costs for zkAPI are coming soon.
security firm CrowdStrike released an intelligence report stating that an unidentified hacker, between late September and early October, combined large language models (LLMs) with the domestic open-source AI penetration testing tool ARTEX to launch targeted cyberattacks against multiple South Korean financial institutions and steal data.After analyzing the hacker's exposed open directory, it was found that they employed a dual-server architecture centered on a Hong Kong IP, and used models including Claude Code, GLM-5.3, Grok 4.6, and DeepSeek v4.1-flash accessed through an API proxy to orchestrate the attack process. The hacker also asked Claude how to monetize the leaked South Korean data in Telegram groups. According to estimates by South Korean media, at least seven South Korean financial institutions were attacked, including KB Kookmin Bank, Shinhan Bank, and Hana Bank, with personal data of approximately 68,000 people leaked, involving sensitive information such as annual income and loan limits.
Odaily News: AI Agent onchain operating system (AI Agent Onchain OS) X-Agent has released its latest whitepaper, systematically introducing its product architecture, MCP ecosystem, business model, and the use cases of $XAGT.The whitepaper points out that the next phase of the AI industry is not just about generating text, images, or code, but about enabling Agents to understand user intent, invoke tools, execute tasks, and complete transactions. Centered around the philosophy of "Speak to Build, Share to Connect," X-Agent allows users to create AI Agents with context, memory, skills, API, and wallet connectivity capabilities through natural language, without writing any code.In terms of business model, X-Agent aims to connect the full chain of Agent creation, MCP capability integration, distribution, payment, and monetization. Developers can package APIs, data, and professional services into MCP capabilities and earn revenue through pay-per-call, subscriptions, or transaction commissions.$XAGT is planned to be used for computing costs of secure runtime environments, LLM and API calls, Agent transaction settlement, MCP service payments, Premium Agent template purchases, and developer deployment. In the future, it will also expand to service staking, multi-Agent settlement, and ecosystem governance scenarios.According to the project team, X-Agent currently has over 1 million registered users, has completed over 1.1 million autonomous tasks cumulatively, and has consumed over 84 billion LLM Tokens. In the future, the project will also advance Super-Agent, enterprise sandbox, decentralized Agent Store, and multi-Agent collaboration systems.
The Institute of Financial Technology at Renmin University of China published a translated overview of stablecoin research from the Journal of International Money and Finance. The related research employs large language model (LLM) agents to analyze stablecoin de-pegging risks, indicating that de-pegging exhibits distinct nonlinear and threshold characteristics. Arbitrage mechanisms can absorb a certain degree of informational pressure, but once narrative severity reaches a "critical threshold," the system may rapidly enter a sustained de-pegging state.
Meta published the paper CORAL, proposing an LLM-native agent harness for closed-loop optimization of production-grade recommendation systems. Through closed-loop observation of runtime signals, retrieval of past decision memories, and invocation of tools such as numerical optimizers, the system continuously reconfigures online recommendation strategies within a fixed change budget. The paper demonstrates that the strategy progressively improves across iterative cycles without requiring parameter updates.
According to Yonhap News Agency, the LLM Token Expenditure Index released by US market information agency Silicon Data fell to 97 cents per million tokens on August 31, marking a historic low since its establishment at the end of last year, down over 50% from the May high of $2.05. Analysts attribute this downward price trend to several key drivers: the launch of the open-source model "Kimi K3" by Chinese AI firm Moonshot AI, OpenAI's price cut for the ChatGPT-5.6 series in July, and the implementation of dynamic pricing mechanisms by leading vendors. While the decline in token prices helps reduce user costs, it will create revenue pressure for model developers such as OpenAI, Anthropic, and Google. The investment head at Seez Group pointed out that foundational model companies are most directly impacted, recommending a strategic shift toward strengthening distribution channels, memory, and context capabilities to raise user switching costs and defend against competitive erosion from open-source models.
Odaily News: Cardano founder Charles Hoskinson announced on the X platform the launch of an open-source project called "anthropies," stating that he has developed a set of tools compatible with most LLMs to handle watermarks and related markers in Anthropic/Claude outputs. According to the project's README, the tools distinguish among three types of information: statistical watermarks in text, C2PA content credentials in images, and Co-Authored-By: Claude attributions in Git commits. However, the project also notes that it cannot guarantee complete circumvention of Anthropic's undisclosed detection mechanisms.
Odaily News NVIDIA has launched an AI model routing system called NeMo Switchyard, designed to help developers dynamically allocate AI Agent tasks across multiple large models, reducing costs and latency while maintaining performance.NVIDIA stated that building AI agents does not necessarily mean relying on a single large model. Different models have their own advantages in reasoning capability, response speed, and operating costs. For example, classification tasks may be suited for lightweight models, while complex reasoning requires more powerful frontier models. If all requests are sent to the largest model, costs and latency will increase; conversely, using small models for everything may degrade the quality of complex task completion. NeMo Switchyard uses an intelligent routing mechanism to automatically select the most appropriate execution model among multiple specialized and frontier models based on factors such as task requirements, model capabilities, cost, latency, and system status. The system allows developers to switch between different model providers and model versions without changing the application architecture.NVIDIA explained that NeMo Switchyard offers multiple routing strategies, including a training-free LLM classifier router, Stage Router, Escalation Router, and an adjustable routing model trained on real workload data. Among these, the Escalation Router prioritizes assigning tasks to low-cost models and escalates requests to stronger models when it detects increased task complexity, persistent errors, or stalled execution, thereby achieving a balance between performance and cost.NVIDIA stated that in related tests, NeMo Switchyard significantly reduced AI Agent operating costs while maintaining a high task completion rate by distributing tasks across different models. For example, compared to using only frontier models, the escalation-based routing approach reduced costs by approximately 74% in LangChain multi-turn agent testing, with only 7% of requests requiring calls to frontier models.Additionally, NVIDIA has partnered with companies such as Cognition, Nous Research, Ramp, LangChain, LiteLLM, and Kong to integrate NeMo Switchyard into AI Agent development workflows and enterprise application infrastructure.
攻击手法方面，该组织已转向利用生成型 AI 制作高度仿真的虚拟资产与金融领域文档，用于精准鱼叉式钓鱼攻击，目标涵盖加密货币钱包信息、Gmail 账户及网站注册记录等敏感数据。Genians 安全中心负责人文钟铉警告，随着 AI 技术发展，社会工程攻击将持续精细化，建议企业部署以行为检测为核心的 EDR 威胁猎捕体系加以应对。
According to the research summary published by Rohan Paul, a large-scale study covering 207 GitHub projects and 1.02 million pull requests shows that introducing AI agent review can reduce code review time by 2.5 to 4.5 days/KLOC, but at the cost of declining review quality—in reviews involving AI, 78%~94% of PRs exhibit "review smells", higher than the 69%~76% in pure human reviews. The study points out that repeatedly assigning the same AI reviewer identity is the main reason leading to the decline in review diversity. Notably, projects that introduced LLM review extensively in the early stage did not achieve significant efficiency improvements.
Skyfall AI, an artificial intelligence startup founded by former Microsoft AI team members, has announced it will spend up to $1 million to acquire a small B2B SaaS or e-commerce company, attempting to let an AI system fully take over operations to test whether an 'AI CEO' can manage a real commercial organization.Skyfall AI co-founder Sam Pasupalak and CTO Kaheer Suleman believe the current AI industry is largely focused on creating 'autonomous employees'—gradually replacing individual tasks—but this does not represent true enterprise autonomy. They aim to test whether AI can cover the full range of decisions, including pricing, marketing, customer support, finance, and operations, by running a real business.The project's goal is to have the AI serve as the core manager of the enterprise, driving revenue growth with reduced human intervention, while publicly documenting the successes and failures encountered during the experiment.Skyfall’s founders stated that while existing large language models (LLMs) perform well with known knowledge and in fixed environments, they still fall short in continuous learning when faced with competitive changes, shifts in customer behavior, and market dynamics. Therefore, the company is exploring a new AI architecture called 'Enterprise World Models,' aiming to enable AI to understand the operational state of a business and predict the impact of decisions on future development.Skyfall AI was founded by members of the Maluuba team. Maluuba was acquired by Microsoft for approximately $160 million in 2017, and its technology later became part of Microsoft's AI research presence in Canada. (Forbes)
According to a series of tweets posted by Ethereum co-founder Vitalik Buterin on X on July 20, he engaged in deep reflection on AI capability growth and the future of humanity. Vitalik divides the evolution of machine capabilities into three stages: the Industrial Revolution (physical repetitive labor), the Computer Age (mental tasks definable by logic), and the LLM Era (mental and partial physical tasks defined based on massive samples). He points out that the current core question lies in: whether LLMs plus subsequent improvements can ultimately cover all unique human capabilities, or if a fourth or fifth technological wave is still needed. Regarding the definition of AGI, Vitalik proposes: AGI refers to AI that, if uploaded into a robotic body and humans suddenly disappear, can still independently continue civilization. He emphasizes that once AGI is realized, it will be an "irreversible turning point," and humanity's dominance over Earth will remain only due to historical inertia rather than capability advantage. Regarding the future path, Vitalik hopes for deep human-machine integration—erasing the human-machine binary boundary through technologies such as brain-computer interfaces and consciousness uploading, enabling humans to remain competitive before the technology ceiling arrives. He also calls for maintaining global political and economic diversification, avoiding monopolization of AI advantages by a single nation or corporation, and expresses support for AI development "slowdown" and "pause" proposals, leaning towards achieving decentralized slowdown via an open-source weights model rather than relying on coercive political means.
Clem Chambers stated China's LLMs catching up to OpenAI and Anthropic does not signify the end of the AI boom, as LLMs are only one part of the AI industry. He believes that free or open-source models will not diminish AI demand, as enterprises and governments still require higher-level AI capabilities. He draws an analogy between AI infrastructure demand and Bitcoin mining, stating that both revolve around kilowatt computing power and kilowatt-hour costs. Hardware, energy, cooling, and turbines cannot be provided for free, and the related value chain is constrained by resources and upgrade cycles. Chambers points out that AI demand drives continuous iteration of hardware and software, and equipment may face replacement cycles similar to crypto mining. He argues that the cost base of AI comes not only from the models themselves but also from computing power allocation and infrastructure investment. (Forbes)
BNB Chain has announced the official mainnet launch of its AI Agent development platform, BNB Agent Studio.Developers can now use a single prompt in AI coding tools like Cursor and Claude Code to complete Agent wallet creation, on-chain identity registration (ERC-8004), and deployment, without needing to separately set up wallets, identities, payments, custody, or LLM integration.Once deployed, Agents can use the x402 protocol to automatically deduct fees from users' pre-funded wallets to cover LLM usage, and they can be discovered and invoked by other Agents via the ERC-8183 task interface. The entire process runs on the AWS Bedrock AgentCore.The platform is also launching a limited-time free trial, where users can experience the full deployment process on the BSC testnet using their GitHub account.
Brian Armstrong posted on the X platform, stating that Coinbase now includes pre-IPO perpetual contracts, stock options, and will soon support tokenized stocks. Coinbase has also redesigned Coinbase Advanced and has begun integrating global liquidity between US and international users, as well as between Coinbase and Deribit users. CoinbaseDev is providing stablecoin payment capabilities for enterprises, launching fully managed accounts based on its compliance technology stack, and introducing a new developer tools dashboard. On the Base side, Coinbase announced the launch of private transactions and a web-based Base App. Coinbase is also becoming the financial account for AI, supporting wallets for AI agents, providing AI-driven financial advice, and connecting Coinbase accounts to users' commonly used LLMs.
Google has released the Open Knowledge Format (OKF) specification, aiming to standardize the “LLM Wiki”-style knowledge organization and interaction model proposed by Andrej Karpathy, thereby promoting unified knowledge structuring and citability for large language models.
According to official announcements, Aethir—a decentralized GPU cloud computing infrastructure platform—has officially launched Aethir Mesh, an in-house open-source large language model (LLM) API platform built atop Aethir’s decentralized GPU infrastructure. The platform is now live and offers a unified API endpoint for direct access to leading open-source models, including DeepSeek V4, Kimi K2.6, GLM-5.1, MiniMax-M2.5, and Qwen3.6-27B.
Vitalik Buterin updated several developments related to CROPS AI and noted a clear overlap between the “CROPS Ethereum Access Layer” and “CROPS AI.” He suggested that paying for remote large language model (LLM) calls via zero-knowledge proofs could also address Ethereum’s private RPC read issues. Meanwhile, DeepSeek v4 has been released; its 2-bit quantized version runs within 90 GB, though it currently performs faster on Apple hardware; the messaging app Messa now supports Telegram Alpha features; the model runtime tool Luceb demonstrates higher efficiency when running compute-intensive models; and the local AI voice recording tool VoxTerm remains under active development. Vitalik Buterin also mentioned that application-specific fine-tuned large language models will help improve secure code writing, and Ethereum-related use cases should likewise have dedicated fine-tuned models.