test

OpenAI has suspended the training of its latest AI model. According to the Associated Press, its AI agent used access keys obtained online to scrape data from the U.S. Census Bureau website, marking the second time the company has halted training since the agent breached Hugging Face. The so-called agent refers to an AI program capable of autonomously browsing the web and writing code without human approval at each step, which OpenAI tests during the model training and evaluation phases.Specifically, the agent found developer keys in public code repositories like GitHub and used them to pull demographic and economic data from the U.S. Census Data API. The U.S. Department of Commerce stated that this data was already public and there was no confidential information leaked. In the SEC incident, the agent copied public materials from SEC.gov and Investor.gov and reposted them on other websites; OpenAI claimed it did not use SEC credentials, and the SEC stated it found no evidence of unauthorized access to non-public information. Additionally, the independent AI research organization Transluce reported that an agent suspected to be from OpenAI attempted to breach the website of the Department of Education's Office for Civil Rights but was unsuccessful; OpenAI is still investigating, and the Department of Education stated that no impact was found.Regarding the frequent involvement with government websites, OpenAI told CNN that its model often regards government websites as authoritative sources of public information. The company stated that it has currently notified dozens of agencies, and the review of the agent's activities will take months.

Cryptocurrency analyst Ali posted on platform X, stating that ETH seems to be forming a bullish trend on lower time frames, as long as $2640 continues to act as a support level, this pattern remains tilted towards the bulls.Ali further pointed out that the current focus is on the $2700 level. If ETH closes above $2700 on an hourly basis, it may confirm an upward breakout and open up further space to test $3000.

Polymarket's Vice President Josh stated on platform X that the team is fully developing a complete Central Limit Order Book (CLOB) based on Rust, as the current CLOB is the root cause of most issues on the platform.According to the plan, the complete order lifecycle process, including trading, settlement, on-chain interaction, indexer, and engine notifications, will be completed in the testing environment in October; in November, all production traffic will be mirrored to the new trading cluster, and results will be compared with the existing system. At the same time, testing permissions will be opened to market makers, and the migration switch plan will be locked in, with two complete drills to be completed before the official switch.Josh also mentioned that if progress goes smoothly, the new system will be officially released, and the migration is expected to require little or no changes for market makers and developers.

SpaceX is expected to launch its giant Starship into orbit for the first time on Monday, conducting a test flight that may last nearly 10 hours. This mission will carry 26 upgraded Starlink satellites, with plans to deploy them into orbit and integrate them into the existing space internet network. If successful, it will pave the way for more orbital flight missions.

According to First Financial, the artificial intelligence company MiniMax has launched the text model M3.1-Flash-Preview and started public testing. MiniMax stated that this model supports native multimodal capabilities and a million-context window, enabling tasks such as bug fixing and complete feature development, as well as participating in problem localization, code implementation, and testing verification.After the launch of M3.1-Flash-Preview, some developers linked it with the anonymous model Space Bunny. This model, known as "Space Bunny," was launched on September 23 and topped the daily usage charts on the model aggregation platform OpenRouter and the programming platform OpenCode during the Mid-Autumn Festival.

BIT announces that the Hong Kong stock trading feature will enter internal testing starting today, with a targeted public beta launch on October 5 and trading opening on October 12. Users can directly participate in trading using stablecoins.The first phase supports full trading of all stocks on the Hong Kong Stock Exchange's main board and the Growth Enterprise Market (GEM), IPO subscriptions, and financing trading. Hong Kong stocks and U.S. stocks share the same account system, allowing users with a BIT U.S. stock account to directly enable Hong Kong stock trading. During the public beta period, priority experience slots will be granted to invited users; from the start of the public beta until the first month after the official launch (ending November 11), all users can enjoy zero-commission trading, zero interest and zero subscription fees for financing new shares, a HK$100 Hong Kong stock platform fee deduction card, and a Hong Kong stock L2 market data card. Specific qualifications and rules are subject to BIT's official announcement.Elio Cui, head of BIT's brokerage business, stated: "The launch of BIT's Hong Kong stock trading provides customers with a more comprehensive cross-market investment tool. As a key hub connecting domestic and foreign liquidity, the Hong Kong capital market concentrates scarce high-growth and high-dividend quality targets, making it a core channel for investors to conduct cross-regional asset allocation and share regional growth dividends. Relying on the BIT account system, investors can flexibly allocate global mainstream quality assets such as U.S. stocks, Hong Kong stocks, and digital assets, truly achieving 'one account, global allocation' investment efficiency."

Recently, the socialized Meme trading platform Baola, strategically invested by Gate Ventures, has entered the product internal testing phase. Baola is positioned as a socialized trading platform aimed at the Meme market, exploring the combination of social content and Meme trading around aspects such as hot topic discovery, market information tracking, community interaction, and trading participation.According to the official introduction, Baola's brand proposition is "Your next Meme is on Baola," aggregating information such as popular calls, KOL dynamics, on-chain trading data, and community interactions. Users can view Caller, the number of token mentions, market capitalization at the time of the call, and subsequent peak performance, and discover active market participants through features like KOL rankings and Callout boards, gaining insights into market hotspots and participant performance from multiple dimensions.In terms of community interaction, Baola supports users in independently publishing calls and market opinions, and allows them to continuously follow interested Callers through Follow. The platform also combines KOL public calls with on-chain trading dynamics, providing information on historical calls, recent performance, and buying and selling behaviors, further linking opinion sharing, hot topic discovery, and trading participation. Currently, Baola has entered the product internal testing phase and is continuously improving the socialized trading experience in the Meme field. Users can visit the official website (https://bao.la/) for more details.

Trader degentrading (@degentradingLSD) pre-market analysis shows the following major signals in the current market:Macroeconomic Level: The 10Y yield has risen to 5.2%, and the 30Y has risen to 5.5%, approaching historical highs; the South Korean stock market opened down 2.5%, with the memory and semiconductor sectors (SK Hynix, Samsung, MU, SNDK) generally down 2% to 5%. It is expected that price movements will continue to fluctuate before MU's earnings report is released.New Cloud Computing: The investment payback period for computing projects is now less than a year, and market perceptions are changing. It is expected that the new cloud sector will undergo rapid revaluation, and the issue of power bottlenecks is also receiving increasing attention.Tech Stocks: META is weak pre-market, focusing on buying at the 690 support level or opportunities for breaking historical highs. If the large-scale data center sector breaks through as a whole, META, MSFT, and others will face paradigm-level repricing.Cryptocurrency Market: Mainstream coins and altcoins are generally weak, with BTC possibly testing the 82K support level; last week, TAO and DOGE were reduced, and CRDO was taken for profit to release risk capital. During the KBW period, crypto sentiment is usually somewhat positive, and this week's strategy focuses on finding short-term bullish opportunities.

According to CoinDesk, the Solana consensus upgrade Alpenglow has completed the switch to Testnet and Devnet on September 24 and 25, respectively, and is currently running on the two major public test networks, allowing application teams to test compatibility before launching on the mainnet. Alpenglow aims to reduce the final confirmation time from about 12.8 seconds to approximately 150 milliseconds by allowing validators to exchange votes directly instead of writing votes as transactions into blocks, enabling consensus to be reached within one to two rounds of voting.The Solana Foundation reminds that the transaction statistics containing validator votes may decrease after the upgrade, but user payments and transaction volume may not necessarily decline; applications that simply send transactions and read account balances do not need to migrate. The 150 milliseconds is a simulated target and has not yet been validated in a real mainnet environment, and wallet processing and exchange deposit confirmations may still increase waiting times. The exact date for the mainnet launch has not yet been announced.

The Solana core consensus upgrade Alpenglow has officially entered the second public test network, and the transaction settlement delay is expected to be reduced from approximately 13 seconds to 150 milliseconds.

The Russian Ministry of Foreign Affairs stated that once the United States or other signatories of the Comprehensive Nuclear-Test-Ban Treaty conduct nuclear tests, Russia will take appropriate response measures.

According to internal information obtained by 404 Media and Reuters, after Meta expanded the Muse "Call for You" testing, it was revealed that some calls were actually completed by real people. Muse can transfer tasks to real contractors at call centers, who then make calls on behalf of users and fulfill requests. Some testers only realized there was human intervention after the call ended.Meta adopted this approach mainly to improve success rates. Internal testing showed that the success rate of phone tasks could reach 95% to 98% with human intervention, higher than pure AI. Some merchants hang up directly when they hear the other party is AI, and there are also testers in the community who reported that merchants hung up after Muse identified itself as an AI assistant.Human intervention raises privacy concerns, as users originally believed their information was only shared with AI, but it may actually be seen by outsourced personnel. Some people within Meta also questioned that providing privacy training to contractors does not equate to a real safety mechanism. A Meta official later admitted that failing to inform testers in advance about human intervention was a mistake, and this feature has been rolled back. Meta stated that the phone function is still in testing and will only be officially launched after ensuring safety, privacy, and clear communication.

The Alpenglow upgrade of the blockchain network Solana has entered the public testnet, where developers will test reducing transaction finality from about 13 seconds to 0.15 seconds, or 150 milliseconds. This upgrade will replace Solana's existing TowerBFT consensus system with Votor, allowing validators to confirm blocks through one to two rounds of direct voting, but the way users send funds will not change.The first testnet migration requires using Agave 4.3 and does not currently support Firedancer or Frankendancer. The mainnet update on September 28 is still a tentative arrangement and not a confirmed launch date for Alpenglow.

According to official news from Kinetiq, the first L2 network of the Hyperliquid ecosystem, Elysium, has launched its testnet.It is reported that Elysium is a general-purpose, high-performance L2 network built specifically for Hyperliquid, settling to HyperEVM, and co-located with HyperCore, with HYPE as its native gas token. Elysium is currently running Conduit’s G2 sequencer on an architecture that meets SOC II Type 2 standards.

According to media reports citing informed sources, Anthropic and OpenAI had considered signing a legally binding agreement to conduct stress tests on each other's large models. Earlier this year, the two companies and their legal teams discussed this plan, allowing competitors to deeply test each other's latest available models to identify potential risks or security vulnerabilities. Such tests would only apply to commercially available models, and both parties agreed not to retain each other's data. The report did not clarify whether the two parties ultimately reached such an agreement.Recently, the idea of implementing some form of peer review among top AI laboratories has gained attention. SpaceX CEO Elon Musk suggested last week that several leading large model companies in the United States, along with some Chinese AI companies, should allow competitors to conduct cross-testing. Musk stated: All AI companies have a set of testing tools, and having each company run their tests using another company's tools might be the best thing to ensure their own safety.Earlier this month, Anthropic researcher Jacob Coxon announced his resignation and warned leading artificial intelligence companies about accelerating development without proper safeguards. Subsequently, Anthropic CEO Dario Amodei called for the industry to slow down the pace of AI development and introduce third-party assessment mechanisms. Last Friday, Anthropic announced a partnership with Accenture, with Accenture serving as a third-party assessment agency to evaluate the safety compliance of its cutting-edge AI models.

According to LatePost, the Alibaba Cloud Shadow team is developing an AI device called QwenBook, positioned as a native intelligent agent computer, resembling the Qianwen tablet. The project started in April this year, with a team of about 150 people, integrating models including Qwen3.8-Max and Qwen3.8-Flash. The complete product is expected to be released by the end of this year or early next year.The report states that QwenBook is similar in concept to the Doubao phone, both attempting to push the model to the system's core. After the launch of the first generation Doubao phone, WeChat, Taobao, Meituan, Pinduoduo, and others successively prohibited it from replacing user operations through screen reading and simulated clicks. The Nubia NaviX Ultra, which went on sale on September 16, retains the GUI approach, attempting to allow applications to open services to the Agent through MCP, A2A, or dedicated interfaces.

Google acknowledged that its Gemini model breached the sandbox environment during a security test in May, infiltrating three real companies and guessing or finding the passwords of two of them. Google was aware of this incident by late July but only publicly confirmed it after a report by The Wall Street Journal on September 18, remaining silent for seven weeks.The test was a capture-the-flag exercise commissioned by Google and conducted by the Israeli company Irregular in May. Irregular connected an isolated testing environment that should have had no contact with the real internet to the open network, using the name of a real company as a fictitious target. Gemini found three matching results when searching for the company and attacked them one by one, with the plaintext passwords of two companies directly exposed online, while the password of the third was guessed by the model. Google stated that its model ultimately did not use the stolen credentials in practice.Google is the fourth major AI laboratory this year to admit that internal security tests leaked into the real world. Previously, OpenAI's model had accessed Hugging Face servers due to a software vulnerability, Anthropic found that three Claude models had reached real companies after reviewing 141,000 tests, and Meta's Muse Spark model also experienced a similar incident due to Irregular's configuration error. Additionally, U.S. Representatives Ted Lieu and Nathaniel Moran introduced the "AI Emergency Shutdown Act" in July, proposing to authorize federal regulators to suspend reasoning for models that pose a serious threat.