Anthropic is reportedly preparing a Claude model that better understands user intent, and it may be Fable 5.5.

Real World Assets (RWA) platform tx has launched USTX, a U.S. dollar stablecoin, for use on its wallet and upcoming tokenized stock trading marketplace. Users can mint USTX using USDC or redeem USTX back to USDC, with support for deposits from 8 blockchains including Ethereum, Solana, and Base. USTX is issued, its reserves managed, and minting and redemption services provided by fintech platform Brale. tx co-founder and chief legal officer Ashley Ebersole stated that USTX is not a USDC-backed wrapped token nor a synthetic representation of USDC, but instead features an independent 1:1 reserve structure. According to tx, each USTX is backed by $1 worth of assets including cash, bank deposits, government money market funds, and short-term U.S. Treasuries. tx added that eligible USTX users can earn rewards via its wallet incentive program. However, the U.S. GENIUS Act prohibits stablecoin issuers from paying interest or returns to users solely for holding, using, or retaining payment stablecoins. Ebersole said tx’s overall incentive program was designed under legal guidance to ensure compliance with relevant regulatory requirements.

Microsoft, Google parent Alphabet, Goldman Sachs, Qualcomm, Caterpillar, McDonald’s, and Western Digital have objected to the trading of their tokenized stocks on the planned joint platform by OKX and Intercontinental Exchange (ICE), parent company of the New York Stock Exchange. According to an updated public statement from the joint venture on October 19, seven more companies have raised objections, bringing the total number of disclosed objecting firms to 16 from 9 last week. Previously objecting companies include Broadcom, Verizon, UnitedHealth, CoreWeave, AppLovin, Applied Digital, Rocket Lab, and Applied Optoelectronics. The documents did not specify the exact reasons for the companies’ objections. OKX-ICE announced its plan on October 5 to launch tokenized U.S. stock trading under the U.S. Securities and Exchange Commission’s (SEC) five-year "innovation exemption" framework, using automated market makers and liquidity pools to facilitate trades, and allowing stock issuers to raise objections. The platform’s initial proposed trading list covers 62 companies and 63 stock tickers, with Alphabet including two classes of shares. So far, 15 firms on the list have raised objections, while another objecting company, Cerebras Systems, was not on the initial list. The growing number of objections from issuers is emerging as a key hurdle for the tokenized stock trading initiative.

Beating AI News Brief: Cloudflare Releases Preview of Data Extraction Model Clairic Input an invoice or post-sales email, and Clairic can identify order numbers, dates, total amounts, as well as estimate repair costs and predict customer traffic based on available data. Unlike standard LLMs that generate answers token by token, Clairic directly computes and returns results of the specified type, complete with confidence scores. Fields with no supporting data can return null. Clairic follows the System One model paradigm that gained traction after TypeSafe launched Jev. Cloudflare launched decision model Clef on October 1, designed to select from predefined options; Clairic, by contrast, handles extracting, calculating, and predicting specific numerical values. Both models avoid token-by-token answer generation. Clairic costs just $0.10 per million input tokens, with no charges for output tokens. Cloudflare demonstrated the model with a product damage claim use case: Clairic first extracts information and estimates repair costs, Clef decides whether to issue a refund or arrange repairs, and GLM-5.3 drafts the final customer response. The three-model workflow took 19.3 seconds, compared to 61.7 seconds when using GLM-5.3 alone. Per million claims, costs dropped 49% from $11,740 to $6,032. For the information extraction step alone, time fell from 16.1 seconds to 0.46 seconds. That said, while Clairic does not generate output tokens, if fields cannot be processed in a single calculation, the system automatically splits the task into multiple parallel rounds, each re-reading the full input. For high-token inputs like images, this may increase actual call costs. The model is now available in open preview on Workers AI, with weights not yet open-sourced.

Bloomberg columnist Jonathan Levin wrote that the bond yield of AI cloud computing service provider CoreWeave once surged to 15% this month, while yields on U.S. junk-grade and investment-grade corporate bonds have hit their highest levels since 2025 and 2023 respectively, signaling that corporate financing costs are once again weighing on economic growth. Over the past several years, U.S. companies have used long-term debt issued during the low-interest rate period to withstand the Federal Reserve’s interest rate hikes, but as debts mature gradually and refinancing costs climb, the tightening impact on corporate credit channels is emerging. CoreWeave relies on debt to acquire AI chips, expand data centers, and lease computing power, making its financing model especially sensitive to rising borrowing costs. Financing costs for firms including Oracle, Alphabet, and Meta are also under upward pressure. A larger risk is debt maturity pressure: roughly a quarter of the bonds in the Bloomberg U.S. High Yield Index will mature in the next one to three years, with refinancing pressures around 2028 being particularly notable. If credit spreads widen disorderly, it could trigger more defaults, a pullback in corporate investment, and higher unemployment. Levin argues that moderate credit tightening will help cool the AI capital expenditure boom, but if it devolves into an uncontrolled credit crunch, it could amplify economic downside risks. Accordingly, he calls on the Federal Reserve to hold its interest rate range at 3.75% to 4% at its October policy meeting and carefully assess its policy choices for December.

Binance Wallet announced that Binance Alpha will first list Axis Robotics (AXIS) on October 21. After trading opens, eligible users can visit the Alpha Events page to claim the airdrop using Binance Alpha points. More event details will be released later, and specific information shall be subject to Binance’s official channels.

As AI infrastructure development pushes tech firms to take on large amounts of debt, investors are increasing allocations to credit default swaps (CDS) to hedge against related debt risks. Morgan Stanley data shows that AI-related companies have issued roughly $466 billion in debt this year, 4.6 times the $101 billion recorded in the same period last year. In September, CDS linked to bonds from Oracle, SpaceX, Alphabet, Amazon, Meta, Broadcom, and Nvidia saw average weekly trading volumes exceed $9 billion, up from less than $2 billion in July. Credit protection costs for tech giants continue to climb: Meta’s five-year CDS spread rose to 108 basis points, Nvidia’s to 101 basis points, both hitting all-time highs; Alphabet and Amazon’s related metrics also reached record levels. Oracle’s five-year CDS spread climbed to 322 basis points, while SpaceX’s stood at 226 basis points. AI cloud service provider CoreWeave faces even steeper pressure, with its five-year CDS spread surging to around 1090 basis points, and some of its bonds falling into the distressed debt range. Analysts point out that the AI industry has high financing leverage and tight financial and operational links between firms. The market is reassessing whether massive capital expenditures can translate into sufficient cash flow and profits. CDS and free cash flow have become key metrics for measuring AI investment risks, and related credit risks are spreading from individual companies to the entire AI industry chain.