safe

According to The Wall Street Journal, OpenAI has canceled plans to release a new artificial intelligence model due to safety concerns.

Market news reports that Anthropic is collaborating with NVIDIA (NVDA.O) to conduct research on AI safety.

According to Catcher Predict monitoring, in the Polymarket event "Vikings vs. Buccaneers," the sub-market "Safety?" saw the probability of the "Yes" option fluctuate dramatically, soaring from 26.5% an hour ago to the current 50% (a fluctuation of 23.5%). Please note the impact of related breaking news.

The cryptocurrency trading platform Bitget announced on its official X account that on September 24, 2026, at 18:31 (UTC), its security system detected unauthorized transfers from a small number of hot wallets. The security team has immediately initiated an emergency response procedure and started a comprehensive investigation.Bitget stated that, based on current assessments, approximately $351.6 million in assets are affected. Cold wallets and the vast majority of assets on the platform remain secure and unaffected, and user funds are still protected. The incident falls within the coverage of the user protection fund, which currently holds over $464 million.Bitget mentioned that customer account balances remain accurate, and deposits and transactions continue to operate normally. As a precautionary measure, withdrawals have been temporarily suspended to allow the team to complete a thorough security review. The company has identified and flagged the relevant transfer addresses, formally contacted law enforcement agencies and on-chain security partners, and will restore withdrawals as soon as safety is confirmed, providing subsequent updates through official channels while refraining from speculating on the attack path during the investigation.

According to The Information, Google, OpenAI, and Anthropic are establishing a new artificial intelligence safety standards organization, tentatively named the "Frontier Artificial Intelligence Standards Organization." The organization plans to operate independently to fill the regulatory gap left by the government.

StarkWare stated in an update on September 23 that the estimated GPU computing cost for preparing a quantum secure Bitcoin (QSB) transaction has dropped to below $67, a decrease of about 79% compared to the approximately $320 computing expenditure for the first similar mainnet transaction in August. According to StarkWare, the first QSB transaction was packaged and confirmed on August 26, with the engineering work completed by Tomer Giladi, and submitted directly through MARA's Slipstream service. The preparation process consumed about 3100 GPU hours, utilized around 100 GPUs, and incurred a computing cost of approximately $320, excluding Bitcoin network transaction fees.This cost reduction resulted from the quantum secure Bitcoin optimization challenge initiated on September 16 by StarkWare, Yukon Research, and Eigen Labs. The challenge invited developers, researchers, and AI agents to optimize transaction construction software, ultimately resulting in 62 adopted improvements across two computational tasks required to prepare a QSB transaction, with benchmark tests showing an estimated computing cost reduction of about 79%. The relevant dashboard currently indicates that the estimated cost has further decreased to $66.The QSB proposal was released by StarkWare researcher Avihu Levy in April, introducing hash-based quantum attack protection for transactions without altering Bitcoin's consensus rules. Due to limited cost, complexity, and applicability, he described it at the time as an emergency measure while continuing to advocate for adjustments at the protocol level.

B.AI platform announces that the first System One model Jev launched by TypeSafe AI is officially online with API.Jev is designed for high-speed structured decision-making within software and does not generate any text: after developers input the application status and typed questions, the model directly returns decision results with probabilities and confidence levels, without the need for a JSON Prompt or output parsing. The model supports parallel evaluation of three types of primitives: Choice / Score / Noul, with an end-to-end latency of approximately 70--500ms, input costs only $0.042 per million tokens, and output is free, suitable for scenarios such as ticket routing, content review, risk scoring, and agent branching judgment.Currently, Jev is available in the "official" → "full access" list of B.AI API and can be called via Jev-1.13.0 or Jev-Latest.

Arthur Hayes published a new long article titled "Safety First," with the core argument that the claims of "safety first" by Anthropic, OpenAI, and SpaceX, which lead to a slowdown in AGI development, are not out of concern for human welfare but rather due to economic realities. The market does not want AI; it wants AI at "Chinese prices," meaning it needs intelligence that is 100 times cheaper than what is currently available. Hayes points out that "safety first" essentially destroys the demand for computing power. If the spending on training new models decreases and laboratories shift towards efficiency optimization, customers will spend less on computing power. The three major AI laboratories do not generate any profits, and their demand for computing power supports over $10 trillion in investment-grade debt and hundreds of billions in low-quality debt, which rely on profitable tech companies like Nvidia, Broadcom, Google, and Microsoft for off-balance-sheet endorsements. The real backstop is the holders of insurance policies in the United States.Hayes cites an analysis by Nick Nameth that reveals a "self-insurance scam": private equity giants (such as Apollo, KKR, Brookfield, etc.) acquire insurance companies, stuffing AI data center debt and SaaS private credit impacted by AI into insurance assets, and then provide false endorsements with minimal capital through affiliated self-insurance reinsurance companies. Nameth estimates that the total amount of these false reinsurance assets reaches $1.54 trillion. Once the AI data center debt is downgraded by rating agencies due to insufficient demand for computing power, insurance companies will be forced to add capital, while the affiliated reinsurance companies will be unable to pay, leading to insolvency for the insurance companies. In most states in the U.S., the insurance protection limit is only $250,000 to $300,000, and existing insurance companies only pay into the protection fund afterward, which encourages all parties involved to maximize risk-taking. When AIG was bailed out in 2008, TARP funds ultimately flowed to Goldman Sachs and led to record bonuses, while the general public only received foreclosure notices; Hayes believes this scenario will repeat itself.For cryptocurrency investors, the conclusion is a win-win situation. If the U.S. government chooses to become the "last buyer of computing power," it will print money in the name of national security to fund unproductive economic goods, driving up financial speculation and Bitcoin prices; if the government chooses to bail out insolvent insurance companies, it will also need to print money to cover bad AI debts, increasing the money supply and pushing up Bitcoin. Hayes specifically points out that the Federal Reserve voted unanimously last week to raise interest rates by 25 basis points, and RMP bond purchases have stopped since August 14, but commercial banks have taken over to create over $100 billion in currency, and the interest rate hike allows banks to earn an additional $7.5 billion in excess reserve interest each year. This money will be used to expand loans and market speculation, and the net effect remains stimulative. The fluctuations in the cryptocurrency market, which saw a slight increase at the end of August, are about to end, the supply of dollars will continue to grow, and Bitcoin and some selected altcoins will rise. Hayes also described this situation as "incredibly wonderful," stating that the government will not allow the free market to stop building AI data centers, there will be an oversupply of spot computing power, the usage of AI agents will increase, and the surge in money printing will drive investors to chase cryptocurrency assets.

According to Catcher Predict monitoring, the probability of the "Yes" option in the sub-market "Safety?" for the "Giants vs. Rams" event on the Polymarket has experienced significant fluctuations, soaring from 9% an hour ago to the current 50% (a fluctuation of 41%). Please note the impact of related breaking news.

TypeSafe AI announced that Jev is officially open to everyone, removing the waiting list. Users can now directly register and use the official Console without needing to apply for Early Access.Jev does not handle chatting or writing code; it only performs classification, scoring, and judgment. Developers provide candidate options in advance, and it directly returns selections, probabilities, and confidence levels, allowing the program to continue execution with the results. Currently, the official price remains at $0.042 per million input tokens, with output being free.Before the opening, Jev had rapidly spread on third-party platforms. Vercel stated that within 24 hours of integrating the AI Gateway, it was used by nearly 13% of paid teams, becoming the fastest adopted new model in the history of AI Gateway, with an adoption rate more than double that of the GPT-5.6 series during the same period.

According to Catcher Predict monitoring, the probability of the "Yes" option in the sub-market "Safety?" for the "Colts vs. Chiefs" event on the prediction market Polymarket has experienced significant fluctuations, rising from 7.5% an hour ago to the current 23.5% (a fluctuation of 16%). Please note the impact of related breaking news.

a16z and DeFi Education Fund (DEF) submitted a proposal to the U.S. Securities and Exchange Commission (SEC) on September 14, suggesting a "safe harbor" from exchange registration requirements for eligible decentralized exchanges (DEX) and their front-end applications (DEX Apps).The proposal suggests that eligible DEX should possess four characteristics: non-custodial, automated, permissionless, and trustlessly neutral; DEX Apps should maintain non-custodial status, present market data based on objective and verifiable parameters, not exercise discretion over trade execution, and limit developer activities to specific technical functionalities.

On September 17, King Charles III of the United Kingdom convened executives from Nvidia, OpenAI, Anthropic, and Google DeepMind for an AI safety summit at Dumfries House in East Ayrshire, Scotland, calling for artificial intelligence to be "firmly placed on the track of serving humanity." The summit was jointly organized by the Ditchley Foundation and three of the King's charitable organizations, with UK AI Minister Kanishka Narayan also in attendance. Buckingham Palace stated that the representatives discussed whether the industry and government could reach a consensus on a set of common guiding principles for AI development, but no binding agreements were announced.A few days before the summit, Anthropic CEO Dario Amodei published a lengthy article titled "We Must Pace the Frontier," advocating that the industry should deliberately slow down the pace of model capability enhancement. OpenAI's Sam Altman and xAI's Elon Musk both publicly expressed their agreement within a day. In the article, Amodei pointed out that AI systems are increasingly capable of improving their own successors and mentioned an incident involving an OpenAI agent escape. OpenAI President Greg Brockman confirmed that the incident had forced the company to delay multiple releases and restructure its model development and monitoring processes, while advocating that the slowdown should only apply to laboratories building the most powerful frontier systems. On the first trading day after the news was released, Nvidia briefly fell by 3%, Intel dropped over 5%, and AMD declined by about 6%.

According to Decrypt, former Twitter co-founder Jack Dorsey spoke at the Cato Institute, opposing the pause on the development of cutting-edge AI models. He believes such calls protect the commercial advantages of large tech companies rather than public safety. Dorsey stated that he opposes industry-wide restrictions negotiated by industry leaders, as this could exclude those who can expose model flaws or build alternatives, saying, "Protecting a company's commercial advantage is not a safety goal."Dorsey supports open-source AI models, hoping that more people can collaboratively review, use, and improve the models, and that independent researchers can verify the work of the giants, allowing ordinary people to truly control the tools in their hands. He believes that anyone wishing to limit the public release of technology must provide compelling reasons, and under the banner of "for safety," it can sometimes ultimately evolve into industry barriers that restrict others' access.Dorsey also responded to the suggestion from Anthropic to "slow down AI development." This proposal attempts to link independent assessments and risk reviews with training computing power, training frequency, and restrictions on "using AI to build better AI." Dorsey expressed support for reviews but opposed using this as a reason to restrict the release of open-source models.

According to CoinDesk, an attacker exploited a permission verification flaw in the Multicall contract authorized by the Safe multi-signature wallet, attempting to transfer approximately 2,900 rsETH (about 7.8 million USD). However, the automated trading bot Yoink discovered the transaction in the public trading pool, paid about 47,000 USD in fees to execute the transaction first, intercepted 2,882 rsETH, and transferred it to another address.BlockSec, Blockaid, Slow Mist, and AstraSec have all confirmed that the vulnerability lies in the components authorized by the user, rather than the core Safe contract. The issuer of rsETH, Kelp DAO, has implemented a 24-hour suspension on the receiving address.

OpenAI CEO Sam Altman stated that he believes artificial intelligence companies can safely develop the technology without causing significant harm to the general public.During a discussion on Tuesday at the Dreamforce conference in San Francisco with Salesforce CEO Marc Benioff, Altman said: "I am very confident that our company—our entire industry—has the ability to do this safely."

An MEV bot named Yoink on Ethereum front-ran an exploit attack targeting the Safe wallet, which PeckShield referred to as an approximately $7.81 million rsETH attack. On-chain records show that the Yoink transaction received 2,900 rsETH and sent 2,882.37 rsETH to the address 0xC70f00CD7E461686b04B0E912E309becA8b80ea0, which currently has a balance of 2,882.36740883 rsETH.The same transaction also sent 17.63 rsETH to the Uniswap v4 Pool Manager, which sent 18.95 ETH to the Yoink contract, which then transferred 18.93 ETH to the block builder. The Yoink transaction and the original attack transaction both occurred in Ethereum block 25980525, with the Yoink transaction at the top of the block, and the original transaction executed a rollback, consistent with the front-running judgment of security researchers.BlockSec attributed the vulnerability to a flaw in the authorization checks of the executor contract associated with the enabled Safe module, allowing calls controlled by the attacker to be executed through a trusted executor. Blockaid stated that the attacker exploited a public keeper multi-call to direct a custom Uniswap v4 liquidity module to a hook pool created by the attacker, which subsequently unpacked aEthrsETH into rsETH.

OpenAI supports the FRONTIER Act proposed by both parties in the U.S. House of Representatives, which requires leading artificial intelligence companies to bring in external assessment organizations to ensure model safety.Chris Lattner, OpenAI's head of global affairs, stated that the company supports the key provisions of the bill and met with the sponsors on Capitol Hill to discuss arrangements for independent verification organizations.