Abracadabra proposes an orderly shutdown of MIM and its protocol, with MIM’s effective collateralization rate below 4%.

Beating AI Express News: OpenAI launched the OpenAI Marketplace at its DevDay event. When enterprises sign large-scale contracts with OpenAI, they agree to a minimum spending commitment over the contract term. Now eligible clients can include a portion of their third-party software purchases toward this commitment, rather than spending all funds exclusively on OpenAI’s own products. The first batch of partners includes 32 companies, such as Adobe, Figma, Salesforce, ServiceNow, HubSpot, Harvey, Palo Alto Networks, and CrowdStrike. Enterprises can browse products on the Marketplace but cannot place direct orders yet. Actual purchases are still handled via contracts between clients and partners, with partners issuing invoices, and OpenAI crediting eligible amounts toward clients’ spending commitments. This model is similar to the cloud marketplaces of AWS and Azure. OpenAI can refer its existing enterprise clients to partners, while partners gain an additional sales channel. Sam Altman has positioned the Marketplace as part of OpenAI’s enterprise strategy, aiming to leverage ChatGPT’s user and client base for software distribution.

Dongcha Beating AI News: OpenAI unveiled its persistent agent "Dots" at DevDay, only to find its two most intuitive .com domains are not under its ownership. dot.com currently belongs to xAI, redirecting directly to the Grok Bot; dots.com, a domain more closely matching the product name, is also not owned by OpenAI. The timing of dot.com is particularly notable: public WHOIS records show the domain was transferred in July this year, roughly two months before Dots’ official launch. After the news broke, the community quickly speculated that xAI had preemptively acquired the domain, though no evidence has yet confirmed it knew in advance that OpenAI would adopt this name. The history of dots.com is even more unusual: it originally belonged to U.S. women’s clothing chain Dots. When the company went bankrupt and liquidated in 2014, the domain was sold for $335,000 to an unnamed individual buyer, and its actual holder has never publicly identified themselves since. The upshot? Right after OpenAI launched Dots, its two most natural .com entry points are in others’ hands—one of which even redirects to a competitor.

Cathie Wood’s Ark Invest currently holds direct exposure to Kalshi through three of its ETFs: ARKK, ARKW, and ARKF. The firm forecasts the annual trading volume of prediction markets will reach $1 trillion to $5 trillion in the medium term. Ark Invest stated that although Kalshi remains a private company at present, it hopes to enable its investors to participate in the platform as early as possible.

12:30 PM ET yesterday, Trump held a luncheon with executives from multiple AI companies at the White House. Attendees included Jensen Huang, Elon Musk, Mark Zuckerberg, Alphabet CEO Sundar Pichai, Anthropic co-founder and CEO Dario Amodei, among others. OpenAI was represented by President Greg Brockman. Following the meeting, the White House pushed for companies to sign a voluntary responsibility commitment targeting frontier AI models, making the short-term path for U.S. AI regulation clearer. According to disclosures, the agreement will require companies to establish multi-layered security mechanisms, including internal risk monitoring during model training, independent team reviews, external assessments, and board review of relevant results. Trump stated that existing laws and the enforcement capabilities of the Department of Justice and FBI are sufficient to address some risks, and the government prefers tech companies to take on self-regulatory responsibilities first. House Speaker Mike Johnson noted that such voluntary commitments could serve as a basis for future legislation, though no clear bill text or timeline has been set. Trump also referenced the possible establishment of a roughly 10-member oversight committee, though its membership, authority, and enforcement mechanisms have not been released. For the market, the White House’s decision not to introduce model licensing, mandatory reporting, or new federal restrictions will ease short-term regulatory pressure on large tech firms and the AI infrastructure chain. However, the agreement lacks penalty clauses, unified testing standards, and audit details, and has not yet led to new government orders, subsidies, or capital expenditure plans. Going forward, investors will focus on whether companies disclose specific audit rules and whether Congress will advance this self-regulatory framework into formal legislation.

According to an official announcement, Binance will add 7 bStocks tokenized securities as eligible collateral assets for Cross-Collateral Margin, Unified Account, and Unified Account Pro, effective September 30 at 20:00 (UTC+8). The newly added assets include PDD Holdings (PDDB), Forward Industries (FWDIB), SharonAI (SHAZB), Wendy's (WENB), Adobe (ADBEB), Hewlett Packard (HPEB), and Zoom (ZMB). Qualified users can use these bStocks tokens as margin collateral, and the corresponding bStocks trading pairs will support margin trading. Currently, these assets are only eligible as collateral for the three account types above and are not supported for lending at present.

Beating AI News Flash: OpenAI launched MCP Events at its DevDay, opening ChatGPT Work’s existing event-triggering capability to third-party MCP applications. Previously, some connectors including Gmail, Slack, and GitHub already supported similar functionality, and developers can now integrate this capability into their own MCP apps. Once integrated, new messages, comments, or status changes in the application can directly trigger ChatGPT to execute preset tasks. For example, if a bug is submitted in the product feedback channel, ChatGPT can read the relevant details to prepare fix code, tests, and pull requests (PRs); when new review comments appear in documents, it can directly proceed with revisions based on those comments. The core difference between MCP Events and scheduled tasks lies in their triggering methods: scheduled tasks require periodic active checks, while MCP Events enable external applications to push events directly to ChatGPT via webhooks as changes occur.