BTC $82,856 +1.1%ETH $2,445 +1.6%Fear & Greed 55 Greed

Today at a glance

Tracing and trust dominate: stolen funds move through mixers, AI agent payments go live, and US agencies purge DeepSeek.

3 signals
  • Fund tracingDrift-linked wallets moved 25,700 ETH into Tornado Cash in about two hours, a second mixing round after July, while the DFX pool pays early claimants roughly a penny per dollar.#01
  • AIEMudhra Intermediate CA Compromise#04
  • AI complianceThe DHS order took immediate effect, requiring civilian agencies to remove DeepSeek products and third-party services built on its technology and barring new purchases.#05

Stories are ranked by impact; the first three are the edition highlights. This edition displays 14 of 588 candidates.

#01
CryptoEdition highlight
8.5

Drift-Linked Wallet Sends $54.8 Million in ETH to Tornado Cash

Background

Background

The April 1 exploit drained Drift's Solana vaults after the attacker gained admin powers, introduced a fake token and used it to borrow out real liquidity, with Drift putting the haul at $295.7 million, led by $159.3 million in JLP and $71.4 million in USDC, most of which was later swapped and bridged to Ethereum. In June, Drift said forensic firm Mandiant attributed the attack to UNC6862, a North Korean threat group. Drift's relaunched exchange now runs as Velocity, and DFX recovery token claims opened on Oct. 1 against a pool that held about 3.1 million USDT at launch. Tornado Cash, the mixer used for the transfers, was sanctioned by the U.S. Treasury in 2022 for allegedly helping launder criminal proceeds.

Wallets tied to the April 1 Drift Protocol exploit moved 25,700 ETH — about $61.56 million at current prices, the figure reported in the source body, while the headline cited $54.8 million — into the mixer Tornado Cash in batches over roughly two hours on Tuesday. The addresses had been inactive for about two months before the transfers.

The deposits are the second known mixing round by the exploiter, after about 23,095 ETH was sent to Tornado Cash on July 23 and 24, and they move a large part of the unrecovered haul further from recovery as Drift's DFX recovery pool pays early claimants roughly a penny per dollar of verified loss.

Arkham Intelligence data shows the deposits arrived in batches of 100 ETH — about $237,800 each — plus smaller 10 ETH chunks, sizes that match the mixer's fixed deposit pools. After the transfers, the 71 addresses tracked under the attacker entity still held 81,454 ETH, worth roughly $198.58 million, plus about $561,600 in USDT, USDC, Solana and USDY.

google_news · The Defiant · · 5 sources

Analysis and references

Impact analysis

The transfers push sanctioned-mixer exposure onto Ethereum, where the stolen ETH now sits, raising compliance and counterparty-screening risk for exchanges, bridges and other service providers that may receive the funds. On the Solana side, the slow pace of recovery is reflected in the DFX recovery token, which has been redeeming at roughly $0.0104 per token against verified losses.

References

Tags

#Tornado Cash#Drift Protocol#ETH#Arkham Intelligence#sanctions#Solana

#02
CryptoEdition highlight
8.0

A new centralized laundering service FreedomXMR (FXMR) appeared on Hyperliquid C

Background

Background

Hyperliquid is a standalone Layer-1 blockchain whose state comprises HyperCore and the HyperEVM, both secured by the HyperBFT consensus mechanism; HyperCore hosts its spot markets, where tickers are allocated through auctions. ZachXBT has previously documented centralized laundering services operating on the chain. The 87% illicit-inflow ratio and the fund tracing are the investigator's own conclusions.

On-chain investigator ZachXBT reported that FreedomXMR (FXMR), a centralized money-laundering service that appeared on Hyperliquid Core in August 2026, has more than 87% of its inflow volume tied directly to illicit activity. A manual review of all historical FXMR inflows above $10,000 traced over $20 million to ransomware, exploit, malware and social-engineering incidents, and the supporting spreadsheet was shared with the Hyperliquid team.

The disclosure names Hyperliquid Core as a channel repeatedly used by laundering services, and ZachXBT said the chain has previously hosted operations such as b1exch that exploit its permissionless design. The evidence has been handed to the Hyperliquid team, but no confirmation of action by the team was reported.

The traced incidents include the DCENT Wallet, Fetch.ai and Axelar exploits and an unnamed ransomware payment in September 2026, plus the Gamdom exploit in October 2026. ZachXBT said the FXMR admin sources XMR liquidity from KuCoin with more than 3,000 deposits to the exchange, likely using a swap model similar to Wagyu's; the FXMR ticker was bought in a Hyperliquid auction on July 27, 2026 for 518.28 HYPE (about $28,000).

telegram · investigations · · 4 sources

Analysis and references

Impact analysis

The story puts compliance scrutiny on Hyperliquid's permissionless spot market, where ticker auctions let anyone list an asset whose inflows are now alleged to be overwhelmingly illicit, and it flags KuCoin as the fiat-to-XMR liquidity leg that allegedly funds the service. Exposure runs through exchange compliance and monitoring risk for the involved venues and privacy-asset liquidity channels rather than through any direct protocol loss.

References

Tags

#Hyperliquid#ZachXBT#money-laundering#Monero#KuCoin#on-chain-investigation

#03
CryptoEdition highlightTracking · 2 updates
8.0

FN Brief | Launching After Two Delays, Aero Bids Farewell to the ve(3,3) Economic Model

Background

Background

Aerodrome is the central trading and liquidity marketplace on Base, where users swap, deposit liquidity and earn rewards. Its ve(3,3) model combined vote-escrow tokenomics with the (3,3) value-exchange design popularized by Olympus DAO.

Aero completed its twice-delayed merger launch and replaced its ve(3,3) tokenomics with a revenue-linked model, scrapping weekly voting and the fixed emission schedule while expanding to seven chains. The team said the protocol has earned more than $500 million for holders on Base.

The overhaul changes how incentives and governance work at Base's main liquidity hub: weekly vote-directed emissions give way to a system tied to protocol revenue.

The ve(3,3) design it abandons combined Curve-style vote escrow with Olympus DAO's (3,3) mechanism, in which lockers vote on where emissions go each epoch. Aero's replacement ties token economics to revenue and drops both the weekly voting cycle and the fixed emission table.

telegram · foresightnews · · Single source

Analysis and references

Impact analysis

AERO is the incentive token steering liquidity across Base DEX pools, so shifting emissions from vote-bribe-driven weekly cycles to a revenue-linked formula alters the demand channel for lockers and the way liquidity is directed across venues; the seven-chain expansion widens where those incentives can be deployed.

References

Tags

#Aerodrome#AERO#Base#ve(3,3)#tokenomics#DeFi

#04
AI & Tech
8.0

EMudhra Intermediate CA Compromise

Background

Background

Intermediate CAs are delegated signing layers that issue certificates on behalf of a root CA; compromising one lets the operator revoke that intermediate without invalidating the root or sibling intermediates. Revocation has long been the weak point of TLS: browsers check CRLs or OCSP, and Mozilla has layered on OneCRL for intermediates and CRLite for leaf certificates, but many TLS libraries and embedded clients perform little or no such checking.

eMudhra's intermediate ECC certificate authorities were found to have effectively guessable private keys, reportedly single-byte values (one of them literally "4"), traced to a flaw in the KGC script used on a specific day to create those ECC issuers. Browser vendors responded by revoking trust in the affected intermediates, while non-browser TLS clients that do not use modern revocation tooling remain exposed.

Because the intermediate CAs were publicly trusted, anyone who could guess the keys could have forged certificates for arbitrary domains, and that risk falls unevenly across the ecosystem. A Mozilla engineer said Firefox users were protected automatically within a couple of hours via OneCRL, with Chrome, Apple and Microsoft operating similar tooling, but "for the rest of the ecosystem, there has been little to no investment in revocation."

In Mozilla bug 2083449, eMudhra said its preliminary assessment found "a flaw in the KGC script that was used on that specific day to create these ECC issuers," and that review of the KGC code and HSM keys so far indicates a set of original ECC keys was created onboard as expected plus an approved deviation to its usual KGC. Community reports describe eight affected CAs, and one commenter said the keys were allegedly found at a public repository URL, while noting uncertainty about whether the artifacts show stolen keys or a weak generation process.

hackernews · galadran · · Discussion · Single source

Analysis and references

Impact analysis

The exposure runs through TLS trust rather than through any token's fundamentals: crypto exchanges, custodians, wallet backends and node RPC endpoints that terminate TLS in non-browser clients may not consult OneCRL or CRLite, leaving a window for interception of API and webhook traffic until the affected intermediates are removed from those trust stores. Browser-facing services were largely covered by vendor-side revocation, so the residual risk concentrates in server-to-server and embedded/IoT infrastructure.

Discussion

Commenters reacted with disbelief and derision at the single-byte key material, with one noting they put "compromised" in quotes "because I can't call it that without cringing." A Mozilla engineer's explanation of OneCRL/CRLite remediation and the uneven state of revocation across the ecosystem drew the most substantive attention, and others asked for a qualified account of whether the keys were leaked or merely guessable.

References

Tags

#security-incident#pki#certificate-authority#tls#revocation#eMudhra

#05
8.0

US orders federal agencies to remove China’s DeepSeek AI from their systems

Background

Background

Washington and Beijing are competing for leadership in artificial intelligence, and DeepSeek — a Chinese company founded in 2023 that publishes open-source models such as DeepSeek-V3 and R1 — has been a focal point of that rivalry. Earlier US restrictions covered the Commerce Department and portions of the defense, security and intelligence community. Australia banned DeepSeek products from government systems and devices in February 2025. China's Global Times reported that Beijing vowed to protect its firms' rights following the US move.

The US Department of Homeland Security issued an order, filed Monday and published Tuesday, directing federal civilian agencies to remove DeepSeek products, its affiliates, and any third-party products or services built using DeepSeek technology from their IT systems. The directive took immediate effect and also bars affected agencies from making new purchases.

The order extends restrictions that had previously applied to the Commerce Department and parts of the defense, security and intelligence community to the whole federal civilian workforce and its contractors. DHS warned that employee and contractor use of DeepSeek could expose sensitive information to China.

DHS invoked the Federal Acquisition Supply Chain Security Act of 2018 and cited DeepSeek servers located in China as well as security vulnerabilities identified in DeepSeek's iOS application. The department said the technology could reveal details about US citizens and federal programs.

rss · TechCrunch AI · · 3 sources

Analysis and references

Impact analysis

The directive reshapes procurement for federal civilian agencies and their vendors, so companies that embedded DeepSeek models in government-facing products face compliance and re-engineering costs, while US-based model providers positioned for federal work face a narrower competitive field. For crypto markets, the main channel is sentiment around open-weight Chinese AI models and decentralized-AI narratives rather than any direct exposure, since the order targets federal IT procurement rather than digital assets.

References

Tags

#DeepSeek#US Department of Homeland Security#AI policy#national security#US government#China

#06
7.5

Mastercard opens early access to AI agent payments following pilots with Coinbase, Base and Cloudflare

Background

Background

Mastercard first announced plans for the system in June and has since run end-to-end pilots with Cloudflare, Ant International, AWS, BVNK and Coinbase. x402 is an open protocol that revives the long-dormant HTTP 402 "Payment Required" status code, allowing services to quote a price and a receiving address so software agents can pay directly. Mastercard acquired stablecoin infrastructure firm BVNK in August.

Mastercard has opened early access to a system called Agent Pay for Machines, which lets autonomous software agents discover services, agree to terms and pay for them within human-set spending limits. The company unveiled the milestone at Cloudflare's Connect conference, where Mastercard's Jorn Lambert demonstrated a live transaction.

Mastercard frames the system as a fix for a specific friction: paid online services typically require an account, API key or billing relationship with each provider, which breaks down when an agent selects services on the fly. The launch follows end-to-end pilots with Cloudflare, Ant International, AWS, BVNK and Coinbase.

Agent Pay for Machines builds on the open x402 protocol: when an agent hits a paid service returning a "402 Payment Required" response, it generates a digitally signed payment voucher that the provider verifies before delivering the service, with vouchers batchable for later redemption. In the demo, an agent for marketing agency Hype ran a protein-bar campaign across nine payments to two independent merchants on a single budget, with Coinbase supplying the wallet, Base providing on-chain settlement, and BVNK handling fiat or stablecoin settlement through Mastercard's Global Money Platform; Mastercard is also releasing buyer- and seller-side APIs, a developer toolkit and a testing environment.

google_news · The Block · · 3 sources

Analysis and references

Impact analysis

The design routes agent-initiated commerce toward on-chain settlement on Base and toward stablecoin rails operated by BVNK, so demand from autonomous agent payments would show up first in Base network activity and stablecoin settlement volumes rather than in card volumes alone. Early access status limits near-term scale, and the exposure is to payments infrastructure and settlement venues rather than to any specific token.

References

Tags

#AI agents#payments#Mastercard#stablecoins#x402#Base

#07
Crypto
7.5

Webull to Open Perpetual Futures Trading to US Users in November, Initially Covering Bitcoin and Other Crypto Assets

Background

Background

Perpetual futures are futures contracts without an expiration date that use a funding-rate mechanism to keep contract prices close to a reference asset; they originated in crypto markets and account for a large share of crypto trading activity. KalshiEX LLC received a CFTC order of designation as a designated contract market in November 2020, and Kalshi Klear LLC was registered as a derivatives clearing organization on August 28, 2024, after which Kalshi began clearing its contracts through the affiliate.

Online investment platform Webull (NASDAQ: BULL) will begin offering perpetual futures trading to eligible US customers starting in November, with Bitcoin and other crypto assets among the first products, and plans to widen access before the end of the year. The contracts will be listed on KalshiEX, a CFTC-designated contract market, and cleared by Kalshi Klear, a CFTC-regulated derivatives clearing organization, and can be traded directly through existing Webull futures accounts.

The launch gives futures-approved US retail clients access to crypto perpetual futures through CFTC-regulated venues, a product that has largely traded on offshore platforms for US users. Access is limited at first to customers already approved for futures trading.

Webull said the rollout starts in November with a planned expansion of eligibility by year-end; trading runs through existing Webull futures accounts rather than a new crypto-specific product line. The products are listed on KalshiEX and cleared by Kalshi Klear.

telegram · foresightnews · · Single source

Analysis and references

Impact analysis

The channel is venue and regulatory structure: a US brokerage offering perps through a CFTC-regulated exchange and clearinghouse gives US accounts a compliant alternative to offshore perpetual venues, potentially shifting some retail derivatives flow onshore and intensifying competition among regulated US crypto-derivatives venues. Because access is initially limited to futures-approved accounts and a narrow asset list, any effect on aggregate perpetual liquidity is likely incremental.

References

Tags

#Webull#perpetual futures#CFTC#Kalshi#crypto derivatives

#08
7.5

Strive's Preferred Stock SATS Lists on Nasdaq, Raising $630 Million in IPO With a 15% Annual Fixed Dividend

Background

Background

Strive is a publicly traded asset management firm that repositioned itself as a Bitcoin treasury company in 2025, going public on Nasdaq through a merger and accumulating Bitcoin on its balance sheet; its stated aim is to increase Bitcoin per share and outperform Bitcoin over the long run. Preferred stock, unlike common equity, pays a fixed dividend with priority over common shareholders, and cumulative terms require unpaid amounts to be carried forward.

Strive, a Nasdaq-listed Bitcoin treasury company chaired by Matt Cole, closed an upsized IPO of its SATS preferred stock on Oct. 20, selling 7 million shares at $90 each for $630 million in gross proceeds and listing the stock on Nasdaq. The preferred carries a 15% annual fixed dividend on its $100 stated amount, is cumulative, and is scheduled to be paid each business day starting Nov. 9, subject to board declaration.

The deal gives public-market investors a daily-paid, 15%-yielding instrument issued by a Bitcoin treasury company, and Strive has said proceeds may fund additional Bitcoin purchases alongside working capital and income-generating assets. However, the company has not specified how much of the proceeds will actually go to Bitcoin.

Strive estimated net proceeds of about $608.15 million after underwriting discounts, commissions and expenses; at an illustrative Bitcoin price of $82,500, that amount could buy roughly 7,372 BTC, which would lift reported holdings from 36,818 BTC to about 44,190 BTC and potentially exceed Metaplanet's reported 44,000 BTC. At the $90 IPO price, the $15 annual stated dividend equals about 16.67% of the purchase price, and chief risk officer Jeff Walton said the company carries no debt or maturities and is focused on a single obligation: paying dividends.

telegram · theblockbeats · · 2 sources

Analysis and references

Impact analysis

The deal creates a public-market channel through which equity income demand can convert into potential spot Bitcoin buying by a single issuer, with the size of that demand depending on how much of the roughly $608 million in net proceeds is ultimately allocated to BTC. It also adds a daily-paid 15% yield instrument that competes for income-oriented capital and makes Strive's dividend obligation a recurring claim on its treasury.

References

Tags

#bitcoin#Strive#IPO#preferred-stock#nasdaq#institutional-adoption

#09
Crypto
7.5

Liquid Says Bitcoin Swaps Are Restored While Peg-Outs Stay Frozen

Background

Background

Liquid is a federated Bitcoin sidechain operated by a set of functionaries, with L-BTC issued 1:1 against BTC locked in a federation-controlled multisig wallet. Peg-out is the process for redeeming L-BTC for BTC on the Bitcoin mainchain. CoinDesk reported that bridge nodes were shut down after an incident involving roughly 4,000 BTC, valued at about $320 million at the price cited in the report.

Liquid said it has restored Bitcoin swap trading on its network, but peg-outs remain frozen, leaving users unable to redeem L-BTC for BTC on the Bitcoin mainchain amid an ongoing security incident.

The continued peg-out freeze means service restoration is only partial: users can trade on Liquid, but the path for moving value back to Bitcoin’s mainchain remains closed.

Liquid’s peg-out mechanism converts L-BTC back to BTC on the Bitcoin mainchain; restoring swaps does not by itself reopen that redemption channel.

google_news · The Defiant · · Single source

Analysis and references

Impact analysis

The freeze primarily affects L-BTC holders and venues that rely on Liquid for settlement, because internal swaps do not restore mainchain redemption. Broader Bitcoin market exposure is indirect, running through custody and settlement channels rather than spot BTC trading.

References

Tags

#Liquid#Blockstream#L-BTC#peg-out#Bitcoin swaps#security incident

#10
Crypto
7.5

Coinbase will convert USDT, DAI, PYUSD to USDC on Oct 30, triggering taxable events under German law.

Background

Background

USDT, issued by Tether Limited since 2014, is the largest stablecoin by market capitalization and the most traded token by volume. DAI is a decentralized stablecoin maintained by the MakerDAO organization, while PYUSD is PayPal's dollar-backed stablecoin issued by Paxos. All three target a 1:1 dollar peg, so a forced swap between them is economically close to neutral at the token level but not necessarily for tax purposes.

Coinbase will automatically convert German users' USDT, DAI and PYUSD balances into USDC on October 30. Because the exchange is converting the assets on users' behalf, the swap is treated as a taxable event under German law rather than a neutral balance change.

The policy forces affected German users to hold USDC and, per the source, leaves them with a tax liability attached to a conversion they did not initiate. It also shows an exchange steering stablecoin balances toward its own preferred dollar token in a single jurisdiction.

The conversion is scheduled for October 30 and applies to three third-party stablecoins — Tether's USDT, MakerDAO's DAI and PayPal's PYUSD — while leaving USDC untouched. The source frames the resulting tax treatment alongside Germany's one-year holding-period rule for private crypto holdings.

google_news · Pluang · · Single source

Analysis and references

Impact analysis

The swap shifts a slice of German users' stablecoin balances out of USDT, DAI and PYUSD and into USDC, changing the stablecoin mix on Coinbase's German order books and the float of the tokens involved. It is jurisdiction-specific, so the main transmission channel is sentiment around EU stablecoin consolidation and exchange-level custody policy rather than any systemic liquidity shift.

References

Tags

#Coinbase#USDC#USDT#DAI#PYUSD#Germany

#11
Crypto
7.5

JPMorgan predicts Bitcoin will be the highest-return asset of the next decade, beating US stocks and gold

Background

Background

The report is the 31st annual edition of J.P. Morgan Asset Management's long-term capital market assumptions. JPMorgan has been deepening its own blockchain work, bringing JPM Coin, a token representing dollar deposits, to Ethereum and joining BlackRock, Goldman Sachs and the DTCC in tokenized stock efforts. On the day the report drew attention, US spot bitcoin ETFs took in $526.7 million, their best day since Sept. 22, with BlackRock's IBIT accounting for $665.2 million of inflows.

J.P. Morgan Asset Management's 2027 Long-Term Capital Market Assumptions, released Monday, Oct. 19, included bitcoin for the first time, projecting a 10% compound annual return over the next 10 to 15 years. That tops its forecasts for US large-cap stocks (7.5%), US REITs (8.2%), gold (5.5%) and a classic 60/40 stock-bond portfolio (7%), with an arithmetic mean return of 20.9% and 55% annualized volatility.

Bitwise Chief Investment Officer Matt Hougan framed the inclusion as a sign of traditional financial institutions' shifting stance toward crypto. The forecast comes from the asset management arm of the largest US bank, putting bitcoin on the same table as mainstream assets in a report institutional allocators use as reference.

JPMorgan said that "for the first time, we put a range on bitcoin returns, though we do not yet include bitcoin in our main covariance matrices," and instead built a separate, smaller comparison matrix, citing volatility and unstable correlations. The report also pushed back on a popular bull case, calling the idea that currency debasement will spark a rush into crypto "wide of the mark."

telegram · theblockbeats · · 4 sources

Analysis and references

Impact analysis

The projection enters the model-portfolio inputs that institutional allocators rely on, giving consultants and multi-asset managers a citable capital-market assumption when sizing bitcoin exposure — a channel that runs through allocation policy and ETF flows rather than through spot prices directly. JPMorgan's choice to keep bitcoin out of its main covariance matrices, combined with the 55% volatility assumption, limits how fast that input can translate into mainstream portfolio weights.

References

Tags

#bitcoin#jpmorgan#institutional-adoption#return-forecast#market-structure

#12
AI & Tech
7.5

Meta pauses Muse sharing after researchers find AI chats on the web

Background

Background

Meta introduced Muse as a consumer-facing personal AI agent, promoting it at launch as safe, secure and private. Sharing conversations with an AI assistant is a common feature across major chatbot products, which makes link discoverability a recurring privacy concern for platforms.

Meta said it has paused a Muse feature that lets users share conversations with its AI assistant, after researchers at security firm Claroc found more than 6,800 publicly accessible Muse chat links through public DNS records, Ars Technica reported. Claroc published its findings in a report this week.

The pause is a product-level response to a privacy exposure involving conversations users may have assumed were shared narrowly.

According to the reporting, the links were located through public DNS records rather than through any internal disclosure by Meta. Meta describes the sharing feature as paused, not permanently removed.

rss · The Verge AI · · Single source

Analysis and references

References

Tags

#Meta#Muse#AI privacy#data exposure#Claroc

#13
AI & Tech
7.5

Google open-sources TorchTPU, a native PyTorch back end for TPUs

Background

Background

TPUs are Google's custom-designed AI accelerators, historically programmed primarily through TensorFlow and the XLA compiler, with JAX also targeting them. Running PyTorch on Cloud TPU previously required bridging layers, and Google Cloud documentation describes setups in which frameworks such as Ray provision PyTorch workers per TPU device. Competing accelerator vendors have shipped similar efforts, such as AWS's torch-neuronx for Trainium.

Google has open-sourced TorchTPU, a native PyTorch backend that lets PyTorch workloads run directly on its Tensor Processing Units (TPUs). Google detailed the project on its developer blog, describing an "Eager First" design and XLA integration intended to simplify migrating models to TPU infrastructure.

PyTorch is the dominant framework for model development, so first-party support removes a long-standing barrier that forced teams to bridge through XLA or stay on TensorFlow/JAX to use TPUs. Community commenters framed the release as accelerating the shift away from TensorFlow and widening the set of accelerators PyTorch code can target.

The project was first reported by Reuters in December, and Google has now officially confirmed and detailed it; The Stack reports that multiple Google staff were dedicated to building the backend. Google's developer blog highlights XLA integration and peak-efficiency execution on TPU infrastructure as core goals.

hackernews · ilreb · · Discussion · Single source

Analysis and references

Impact analysis

The transmission channel is hardware substitution in AI compute: a first-party PyTorch path lowers the migration cost onto TPUs relative to Nvidia GPUs, which could gradually affect how training and inference capacity is sourced across cloud providers. For crypto, that matters indirectly through AI-compute-adjacent segments such as decentralized GPU rental and compute marketplaces, whose economics are benchmarked against prevailing accelerator rental rates.

Discussion

Commenters were largely positive. One developer said he had built an open-source libtpu implementation for Tenstorrent accelerators that already runs most of TorchTPU, illustrating what lowering Torch to tile-based accelerators involves; another pointed to AWS's torch-neuronx as the equivalent Trainium effort. Others argued XLA is strong compiler infrastructure and that the release is "the final nail in the coffin of Tensorflow," while one commenter suggested that with hardware specifics living in the kernel layer (Pallas or CUDA), JAX versus PyTorch is now closer to a style preference.

References

Tags

#PyTorch#TPU#Google#open-source#XLA#AI infrastructure

#14
AI & Tech
7.5

OpenAI acquires Orchestra Bio

Background

Background

In April, OpenAI launched GPT-Rosalind, a model built specifically for biology and drug discovery and restricted to vetted U.S. enterprise customers such as Amgen and Moderna, alongside a companion product, Rosalind Workbench, that connects the model to scientific tools and data. The deal follows Anthropic's roughly $400 million purchase of Coefficient Bio in April, as rival AI labs compete to own the workflow behind new medicines. OpenAI CFO Sarah Friar wrote in a January blog post that licensing, IP-based agreements and outcome-based pricing will share in the value created as AI moves into drug discovery.

OpenAI announced Tuesday that it has acquired Orchestra Bio, a San Francisco startup whose AI agents track and coordinate drug development programs, and will fold its team and platform into Rosalind, OpenAI's suite of AI models and tools for biologists and chemists. Financial terms were not disclosed.

OpenAI said the goal is to help scientists make discoveries and manage the work needed to turn them into treatments that reach patients sooner, and that the combined team will build agents connecting scientific findings with the decisions and follow-up studies required to turn a promising result into a medicine.

Orchestra's agents map how setbacks such as a failed manufacturing batch ripple through a drug program and compare possible responses, while scientists and program leaders keep the final decision. Founded in the fall of 2022, Orchestra raised a seed round from Andreessen Horowitz, Definition Capital, Insight Partners and 8VC totaling about $20 million; it is unrelated to Nasdaq-listed medical device maker Orchestra BioMed (ticker OBIO).

rss · OpenAI Blog · · 3 sources

Analysis and references

Impact analysis

The acquisition has no direct crypto-market transmission path; its read-through is to sentiment, reinforcing the AI-expansion narrative that traders apply to AI-themed tokens and lending a marginal tailwind to decentralized-science projects that pitch themselves as alternatives to centralized pharma R&D.

References

Tags

#OpenAI#Orchestra Bio#acquisition#life sciences#drug discovery#Anthropic