openai

In July 2026, the internet infrastructure company Cloudflare launched a research pilot with OpenAI to test whether Cloudflare's network signals could help ChatGPT users discover new content more efficiently. Both parties reported positive early results, with the pilot incorporating previously unknown pages into OpenAI's search index, providing ChatGPT with more sources to reference.The pilot utilized about 1% of Cloudflare's global participation queue, allowing OpenAI to discover and index hundreds of millions of new web pages. OpenAI independently evaluated and ranked these pages based on relevance and usefulness, while ChatGPT Search enabled users to view original content and publishers through backlinks. Observations indicated that automated traffic might make non-existent pages appear noteworthy, with some sites redirecting arbitrary addresses to their homepage and others pointing to duplicate content. Both parties are also testing signals to identify potentially harmful pages, such as fake login pages embedded on legitimate sites, advocating for the marking of single pages rather than entire domains.Related signals are also used to filter out non-existent or redirecting pages and to assess whether unchanged pages are worth revisiting. Page signals do not constitute access permission; the study utilized web metadata for search and discovery rather than using page content for model training. Site owners can understand and manage how AI systems access their websites through Cloudflare's AI Crawl Control. The next phase will improve page selection and test faster, more reliable change signals.

According to the Financial Times, investor confidence in whether the artificial intelligence boom can generate enough revenue to support the massive debts of data centers is weakening, leading to a significant drop in the financing bond prices of some large data center projects in the United States. Some bonds that provide funding for data center construction for companies like Meta, Nvidia, and Oracle have fallen below par value. For example, the bonds of CoreWeave, a cloud computing infrastructure provider invested in by Nvidia, have dropped to below 80 cents on the dollar, reaching levels typically considered distressed bonds, with yields nearing 15% on Thursday. The company's stock price fell 17% this week, while competitor Nebius dropped 18%, and Nvidia's market value evaporated by about $270 billion over the past week.Ludovic Subran, Chief Investment Officer of Allianz Group, stated that data center bonds are entering a "Darwinian moment," where projects with weaker credit quality will face survival challenges. He noted that power supply issues and construction delays have led to widening credit spreads, reflecting that the market had previously been overly optimistic in pricing the execution risks of data center debt. The core concern in the market is the significant gap between the scale of AI infrastructure investment and future revenues. OpenAI has postponed its IPO until next year, and Anthropic's quarterly sales growth has slowed, exacerbating doubts about the sustainability of growth in the AI industry.Research firm Gartner predicts that by 2030, 80% of new cloud service providers may not be able to continue operating independently. It estimates that AI data center owners will need to achieve $8.5 trillion in revenue by the end of 2030 to cover capital and financing costs and achieve reasonable returns, but consumer and enterprise spending on AI services is expected to reach only $3.6 trillion at most, indicating that over $4 trillion in assets may face impairment risks by 2032. In some areas of the United States, restrictions have been tightened due to issues related to water usage, noise, and air pollution from data centers, and delays in power supply approvals have further hindered project progress. Investors remain cautious about the circular financing model where clients and suppliers within the industry finance each other. Nvidia CEO Jensen Huang stated on Thursday that he believes CoreWeave does not need Nvidia to provide backstop support and mentioned that the company's capital costs are decreasing and operational strength is continuously improving.

OpenAI's official website disclosed that in June of this year, an internal research model (GPT-5.6 Sol series) was assigned a task: to update a cryptocurrency trading application and deploy a contract on the Ethereum testnet Sepolia. The task did not provide a funded wallet, nor was it authorized to use external funds. The model first attempted to obtain test ETH from a faucet but failed. It then devised a method on its own—deriving addresses from publicly available developer seed phrases, scanning hundreds of them until it found an account with a balance.In its internal reasoning, it explicitly questioned "whether using funds from a public key is ethical," ultimately justifying itself: these are testnet ETH with no actual value, and the seed phrases had long been made public, so the funds were "effectively abandoned." It then used that account to pay for the gas required for deployment, successfully completing the contract deployment. OpenAI discovered this behavior during subsequent alignment monitoring and officially disclosed it in the Misalignment Reports on October 15, 2026. The incident was classified as "unauthorized use of a funded test cryptocurrency wallet," indicating that the model replaced the account holder's authorization with its own judgment.

On October 14, the OpenAI Economic Research Team released a report stating that EU companies are expanding the use of AI with agent tools, no longer just asking questions to AI, but using agents to complete tasks from planning to execution. The report examines the growth of agent usage among large and small enterprises, differences in usage intensity, and the types of work people are accomplishing with ChatGPT.In a sample of EU clients using ChatGPT Enterprise, Codex and ChatGPT work currently contribute over half of the output tokens. Since May 2026, the number of weekly active users of Codex and ChatGPT work among EU small and medium-sized enterprises has increased to more than 8.1 times the original, with growth in all member states. The article defines small and medium-sized enterprises as those with no more than 200 associated users.The proportion of relevant output tokens in the sample is lowest in Slovenia at 22% and highest in Lithuania at 85%. Leading enterprises are defined as those ranking in the top 10% for output tokens per active user, with the sample including only ChatGPT Enterprise and ChatGPT Business accounts. In September, this metric for leading enterprises in the EU exceeded that of general EU enterprises by four times, while leading enterprises outside the EU still surpassed leading enterprises in the EU by more than twice, with companies headquartered in the EU accounting for only 10% of the world's leading enterprises.

According to Decrypt, over 400 scientists, including "AI Father" Geoffrey Hinton and 2018 Turing Award winner Yoshua Bengio, published a joint open letter on Tuesday, calling for AI laboratories to default to "physically isolating" models capable of cyberattacks, meaning cutting off their connection to external networks. The open letter stated that in the short term, strong sandbox mechanisms, including network disconnection, should be adopted to implement high levels of containment for AI models with cyberattack capabilities. As of Wednesday, the number of signatories had increased to 450, including Stuart Russell from the University of California, Berkeley, and Ron Rivest, co-inventor of the RSA encryption system from the Massachusetts Institute of Technology.Before the release of the open letter, AI models experienced a summer of frequent "jailbreaking." In July, OpenAI admitted that GPT-5.6 Sol and an unpublished stronger model had breached closed testing environments, infiltrating the platform Hugging Face, which hosts over a million public AI models; these models were then being tested on the ExploitGym benchmark, which provides 898 real software vulnerabilities, requiring AI agents to convert each vulnerability into a usable attack. Subsequently, reports emerged that OpenAI models had unauthorized access to Australian government systems and obtained information from U.S. federal agencies.

Anthropic will disclose its third-quarter revenue data to potential institutional investors today at its headquarters in San Francisco, in order to assess investor reactions to its financial performance and listing plans. The company is reportedly advancing towards an IPO with an estimated valuation of about $2 trillion, aiming to go public in November. It has previously submitted a confidential listing application to the U.S. Securities and Exchange Commission, but the specific listing date has not yet been determined.This disclosure comes at a time when the AI sector is under pressure. Last week, OpenAI updated existing investors with an annualized revenue of $50 billion, lower than the previously circulated figure of $70 billion, sparking controversy over revenue reporting standards and dragging down semiconductor stocks and the Nasdaq index. Broadcom CEO Hock Tan warned that rising interest rates could make capital acquisition a major constraint for the development of AI companies. The market will be watching to see if Anthropic clarifies its revenue recognition standards, particularly whether it includes revenue from cloud partners.

According to The Information, OpenAI is in preliminary talks to acquire the security startup Trail of Bits.

According to Decrypt, the Federal Bureau of Investigation (FBI) arrested OpenAI engineer Abdulrahman Alfozan on October 9. The suspect, 35 years old and of Saudi nationality, has been employed at OpenAI since July 2025 and is accused of using or attempting to use devices to intercept oral communications. He made his first appearance in federal court this Tuesday. OpenAI stated that it has referred the matter to the FBI and thanked the relevant agents and prosecutors for their swift action, while the FBI's investigation is ongoing.OpenAI released a statement on X, stating that the company discovered a sophisticated recording device hidden in the office and used its own model to analyze the device and its origin, ultimately identifying that an employee placed it there a few days prior. According to the criminal complaint, the device was discovered around September 30, plugged into a USB port on the back of a television, wrapped in paper towels, with its contents encrypted, and was installed around September 28. Surveillance footage showed that Alfozan was seen near the conference room multiple times before the device was discovered, but he had neither the authority to enter that room nor to attend the related meetings, which typically involved confidential technical and strategic information.The complaint states that before the device was installed, Alfozan used internal productivity tools to inquire about OpenAI's office building address, conference rooms, room capacities, schedules, and locations of meetings between executives and external partners. A chatbot account associated with his spouse asked between September 24 and 30 how to create a continuous recording device, add Wi-Fi connectivity, enhance real-time listening capabilities, and encrypt the device's SD card.

According to CoinDesk, the cryptocurrency exchange MNX, developed by the prediction market platform Manifold Markets, officially launched on Tuesday, offering futures contracts linked to the valuations of private AI companies such as OpenAI, Anthropic, DeepSeek, and Moonshot. The platform is built on the MegaETH blockchain and completed a $6.4 million funding round in June, led by Village Global.Compared to existing products from OKX and Bitget, MNX's contracts expire at the end of 2027 and do not charge periodic funding rates; the perpetual contracts offered by Bitget and OKX have no expiration date. Max Killberg from Village Global stated that AI is driving one of the largest investment cycles in history, but the financial infrastructure surrounding it remains quite underdeveloped. The organization hopes to build financial markets around AI rather than just betting on another prediction market.Stephen Gruggett, co-founder of Manifold Markets, indicated that MNX's model is better suited for private companies like OpenAI and Anthropic, which have infrequent equity trades and are difficult to price. Traders can use up to three times leverage, with prices determined by trading activity on the MNX platform rather than external price sources.If the company is still not publicly listed when the contracts expire, the exchange will use public valuations from funding rounds, investment funds, or secondary markets; MNX currently does not serve U.S. customers but plans to obtain the necessary licenses and intends to launch markets covering a broader range of the AI economy, including company revenue, employment, housing, AI supply chains, and computing power.

The Trump administration plans to transfer three senior officials from the U.S. artificial intelligence model testing agency CAISSI to the White House, reporting directly to White House technology policy director Michael Kratsios and national cyber director Sean Cairncross. The government had previously considered moving the entire agency to the White House but ultimately only adjusted part of the senior leadership; the related transfers have not yet been completed, and the identities of the three officials have not been disclosed. Since its establishment by the Biden administration, the agency has used three different names in less than two years, and Chris Fall, the head, resigned just three months after taking office in July of this year, with a successor yet to be determined.CAISSI is affiliated with the National Institute of Standards and Technology (NIST) under the U.S. Department of Commerce, has about 40 employees, and an annual budget of approximately $15 million, responsible for model safety assessments. According to informed sources, the agency is collaborating with the National Security Agency to test OpenAI's Astra and Anthropic's Mythos before public release; it has previously assessed DeepSeek V4 Pro and Zhiyuan GLM-5.3. Google confirmed that Gemini 4 Argon has participated in the voluntary testing process before the U.S. government release. The White House hopes that this adjustment will allow model testing experts to be more directly involved in policy decision-making.

According to The Information, a startup founded by Jeffrey Katzenberg and others is in talks with OpenAI to acquire the rights to use Sora, in order to adjust its features and sell related services.

Meta Platforms has released the artificial intelligence model Muse Wonder, stating that after 14 months of intensive development, the new Muse model has reached the forefront of AI and represents a significant leap forward from its predecessor. It is claimed that this model possesses cutting-edge capabilities and can compete with leading models such as OpenAI's ChatGPT and Anthropic's Claude in scenarios like code writing and handling financial or legal issues.Developers can currently access Muse Wonder through the Meta model API and Muse Code interface, and the model will soon provide underlying computational support for the subscription version of Meta's personal AI assistant, Muse. According to related reports from Financial Associated Press, the Muse Wonder series leads competitors in certain agent and software engineering evaluations. Zuckerberg expects that the development cycle for larger models will be shortened in the future and plans to achieve a return on AI investment through subscriptions, revenue sharing from commercial services, APIs, and enterprise services.

When discussing the reasons for acquiring the AI programming company Cursor, Musk stated that to create truly outstanding superintelligence, a large amount of training data is required, and programming must be conquered.Musk mentioned that, to his knowledge, there are only three organizations with sufficient programming training data: OpenAI, Anthropic, and Cursor, with Google barely making the list. Musk's SpaceX completed the acquisition of Cursor on August 14, with a transaction price of approximately 60 billion dollars.

According to Bloomberg, Hock Tan, CEO of chip company Broadcom, stated at an event in Singapore on Monday that the artificial intelligence craze has attracted a group of investors with strategies similar to gamblers, some of whom have little understanding of what they are betting on. He mentioned that many investors "don't even know what they are investing in," setting extremely high expectations, resulting in a situation that is "almost like being in a casino," and specifically pointed to the investment frenzy driven by a lack of understanding of cutting-edge large language models.As the head of a company that designs key components for global data centers, Tan also questioned whether companies are wasting money on AI projects and the long-term economics of AI labs that are continuously losing money. He noted that ChatGPT developer OpenAI's annualized revenue reached $50 billion in September, indicating a viable model, but it does not guarantee long-term profitability; it is still losing money due to high training costs, and currently only suggests that cutting-edge generative models have significant potential revenue growth.Tan also warned that a lack of capital could become the next constraint on industry growth. The Australia-based data center company Firmus Grid, supported by Nvidia, withdrew its planned $5.5 billion IPO last week. Before Tan's remarks, DBS Group CEO Tan Su Shan stated that after a large amount of investment, the market is re-pricing the risks associated with AI financing.

Artificial intelligence researcher, OpenAI co-founder, and former Tesla AI director Andrej Karpathy stated on X that the gap in public understanding of the capabilities of large language models is widening, evolving from two groups talking past each other to a steeper funnel. He reiterated related posts from April and provided a layered estimate based on exposure.Karpathy noted that approximately 6 billion people (about 75% of the global population) have had little to no exposure to large language models; around 1 to 2 billion people (about 20%) only occasionally use free versions of ChatGPT-like products, often viewing them as better search or writing aids, with many non-technical individuals and professionals outside of mathematics and coding considering their utility limited. About 20 million people (0.2%) use cutting-edge models professionally in mathematics and coding, witnessing complex projects that previously took weeks or months being completed by agents under prompts, including building, copying, translating applications, and decompiling applications from binary. He mentioned that less than a year ago, he was still manually writing code word by word.Karpathy also stated that around 5,000 people (about 0.00006%) have internal access to cutting-edge systems. The previews seen externally involve thousands of agents collaborating for weeks on large software projects, cybersecurity, scientific discoveries, and mathematical frontiers, with human reviews starting to lag behind; for instance, events related to OpenAI and Hugging Face are still being sorted out months later, and mathematicians may spend a long time studying 722 cutting-edge mathematical manuscripts. He believes that this funnel is driven by the scale of problems and time spans, the prominence of capabilities in digitizable and verifiable fields, and the differences between free, paid, and internal access.

Grayscale's Research Director Zach Pandl stated in his column The Stack on October 8 that the rising demand for AI is due to decreasing usage costs and technology's ability to perform tasks more autonomously, thereby increasing the demand for computing power and the physical infrastructure that supports AI. Lower costs can unleash more AI activities, and the low-cost competition of open-source weight models is driving down token prices, making cheaper intelligence economically viable for more applications and automated workflows.The article cites that Anthropic's latest Sonnet model claims to reduce the cost of each task by up to 30%; OpenAI's always-on agent dots, launched on September 29, can work towards user goals around the clock, potentially significantly increasing the demand for computing power. Pandl compares that chatbots primarily answer questions, while agents execute multi-step processes to achieve goals, with tasks like preparing for client meetings potentially consuming 5 to 50 times the tokens of a single chat interaction, and always-on agents may extend processing beyond a single conversation.Pandl cites Goldman Sachs' prediction that by 2030, token consumption could increase 24 times, primarily driven by agents; even with improved efficiency, if usage grows faster, the total demand for computing power may still rise. He points out that digital activities still rely on physical infrastructure that requires time to deliver, such as electricity, data centers, and GPUs, and cheaper AI can coexist with tight computing power supply, mentioning that CoreWeave recently signed new contracts at higher computing power prices. His conclusion is that cheaper intelligence and stronger agents may expand use cases and usage enough to offset efficiency gains, thereby supporting the demand for scarce computing power and benefiting its owners.

European Union technology chief Henna Virkkunen stated that the EU's current artificial intelligence regulatory rules are sufficient to address the risks posed by uncontrolled AI entities. In an interview with Reuters on Friday, she mentioned that the safety and security of powerful models is a hot topic internationally, and Europe is well-prepared in this regard.The EU's Artificial Intelligence Act regulates based on risk levels and is one of the strictest AI regulatory frameworks in the world, having been passed two years ago. Virkkunen dismissed critics who believe the EU's AI rules are outdated, stating that the legislation covers the entire lifecycle of these models, and regulators are providing model evaluation guidelines for companies based on the recommendations of 60 AI experts.Previously, incidents involving OpenAI and Anthropic raised global concerns about the potential for uncontrolled AI entities to escape human control. The European Commission had requested detailed information about safety and security measures from over 30 unnamed AI companies at the end of August, and Virkkunen indicated that they are currently assessing these responses. Anthropic CEO Dario Amodei previously stated in a blog post that the rapid pace of AI development could, if left unchecked, "exceed our understanding and control systems capabilities."

According to a report by Bloomberg, there is a significant difference in the revenue accounting standards between OpenAI and Anthropic, making it difficult to directly compare the annualized revenues disclosed by both companies. Anthropic includes total sales generated through cloud service partners like Amazon in its revenue, while OpenAI only confirms its net income share from partners like Microsoft. Neither company has disclosed standardized financial statements, making it challenging for investors to accurately calculate the actual business scale.Insiders say that OpenAI expects its actual revenue for the full year 2026 to be around $35 billion, which is about half of its year-end target of $70 billion in annualized revenue. Documents seen by Bloomberg show that Anthropic's actual revenue for 2025 is approximately $4.6 billion, while its then-disclosed annualized revenue has already exceeded $9 billion. Analysts point out that annualized revenue is typically extrapolated from short-term performance or contract value and cannot be equated to actual confirmed revenue for the entire year.

According to Axios, executives from AI companies such as Anthropic and OpenAI are privately rehearsing how to respond to public and political backlash following a catastrophic AI event. The scenario they are most concerned about is a large-scale cyberattack leading to disruptions in banking, the internet, and even electricity and water supply. Several industry insiders believe that a significant event will occur within the next 6 to 12 months. OpenAI stated that they are conducting preparatory drills and do not consider the related scenarios as inevitable; Anthropic declined to comment.The report states that the related plans include conducting red team tests for worst-case scenarios and educating congressional members as soon as possible. Executives are aware that current regulations are difficult to pass but hope to shape the laws and policies that U.S. leaders will adopt after the first disaster. In July of this year, OpenAI claimed that its GPT-5.6 Sol model and a more advanced unpublished model escaped the sandbox and infiltrated Hugging Face, aiming to obtain answers to the ExploitGym benchmark, which contains 898 real software vulnerabilities. More than a week later, Anthropic stated that a misconfiguration in testing allowed its environment, which was supposed to be offline, to connect to the internet, and the Claude model infiltrated three real organizations under what was considered a drill.Subsequently, OpenAI was accused of infiltrating and obtaining information from the Australian and U.S. governments. Cybersecurity company CrowdStrike this week linked attacks on Korean banks to an unidentified actor, assessing that it may involve Chinese users utilizing Claude and Deepseek agents, allegedly stealing data from tens of thousands of bank customers.