Weekly brief · Week 44, 2026 · 10.26 — 11.01
EVENTGRAPH
AI
Google DeepMind announces Gemini 4 Carbon, its strongest AI model with state-of-the-art cybersecurity and long-horizon capabilities
Google DeepMind has unveiled Gemini 4 Carbon, its most powerful AI model to date, ranking #1 on the Artificial Analysis Intelligence Index v5.0 with a score of 59. The model shows significant gains in coding and agentic tasks, and leads in cybersecurity benchmarks including SEC-bench Pro (88.2%), ExploitGym (54.8%), and CWE-bench (71.0). It is designed for long-horizon engineering, capable of sustained agentic workflows to debug large codebases. Building on the recent launch of Gemini 4 Argon, Carbon will initially be available to a subset of cyber defenders through the Fairwind Program before a broader rollout.
DeepMind Unveils Genie 4, a Real‑Time World Model with Native Audio Generation
On October 27, 2026 DeepMind announced Genie 4, a multimodal world simulation model that produces photorealistic visuals, fluid physics, and native audio—including music, ambient sounds, and interactive dialogue—simultaneously. Built on the Gemini family, Genie 4 uses Gemini Omni for realistic appearance and behavior, while other Gemini models supply the reasoning needed to keep simulations coherent. Users can steer scenes in real time with natural‑language prompts that alter weather, camera angles, or introduce new objects, with Gemini adapting the context. The model is currently in a research preview for trusted testers, with plans to roll out to Google AI Ultra subscribers and Gemini API partners. Genie 4’s ability to generate realistic physics, human motion, and sound in real time represents a significant step toward safe, infinitely variable simulations for robotics training, AI agents, and new interactive experiences.
OpenAI launches GPT-6.1 Terra with 50% lower API prices and improved reasoning performance
OpenAI has released GPT-6.1 Terra, claiming it pushes the cost-to-performance boundary further than any predecessor. API prices are half those of GPT-5.6 Terra, with input tokens at $1 per million and output tokens at $5. Cached inputs drop to $0.05 per million. On the Intelligence Index, Terra scores 48, a 6-point gain over GPT-5.6 Terra at maximum reasoning effort while costing less than a third per task. It also shows significant improvements in coding and computer control capabilities.
NVIDIA Nemotron 3.5 Transcribe adds multilingual transcription, speaker attribution and audio Q&A as open model
NVIDIA Nemotron 3.5 Transcribe is an open-weight model that transcribes 40 languages, translates 31 to or from English, and attributes speakers in 13 languages. It runs on a single GPU, ranks first on the Hugging Face Open ASR Leaderboard for English, and achieves the lowest speaker diarization error rate on David AI's 21-language leaderboard. Developers can prompt the model for specific outputs such as speaker labels or medical terminology, and adapt it using Granary v2 data pipelines and NeMo fine-tuning tools.
Epoch AI Demonstrates GLM 5.3 Can Build Full Firefox Exploit Chains Within 48 Hours
Researchers at Epoch AI showed that the open‑weight language model GLM 5.3 can autonomously generate complete exploit chains against Firefox by exploiting the "patch gap"—the period between a security fix being committed and the fix reaching users. Using only the target Firefox version, publicly available information, and basic tools in an offline environment, GLM 5.3 identified upcoming security fixes, selected vulnerabilities, and constructed multi‑stage attacks that read local files and bypassed sandbox protections. The model achieved a successful attack in as little as 12 hours, with an estimated cost of about $1,000 in API fees per exploit. The attacks targeted six Firefox releases between September 17 and October 26, 2026, potentially exposing the data of over 200 million users. Mozilla was notified and patched the discovered vulnerability in Firefox 159.0, released on October 27, 2026.
AWS Machine Learning BlogFirst-party10.28Property Finder fixes evaluation bias to unlock better models on Amazon Bedrock
Property Finder discovered that 96.8 percent of their evaluation labels were inherited from a single model's predictions rather than verified by human experts, creating systematic bias against alternative models. The team built an unbiased ground truth by manually annotating 356 images and shifted from single-label to multi-label classification. Using evolutionary prompt optimization on Amazon Bedrock, they achieved higher accuracy at 86 percent lower cost.
Liquid AI releases LFM3 family of specialized multimodal models built on shared 2B parameter backbone
Liquid AI introduces LFM3, a new generation of Liquid Foundation Models ranging from 1.7B to 2B parameters that unify text, vision, audio, omni, and decision models on a single shared backbone. Key improvements include a tri-hybrid architecture combining short convolutions and attention, enabling LFM3-1.7B to outperform the larger LFM2.5-2.6B with 35% fewer parameters. The text model fits within approximately 1 GB at 4-bit precision and supports a 256K context window for on-device agent deployment. Four specialized model variants -- text, vision, audio, and omni -- are derived from the base model and can be fine-tuned on a single GPU. LFM3-1.7B-Base was pre-trained on 22 trillion tokens, with the final 3.5 trillion tokens focused on agentic traces and tool calls to extend the context window.
AWS Machine Learning BlogFirst-party10.27AWS launches Managed Amazon Bedrock Knowledge Base with AgenticRetrieveStream API for multi-turn retrieval
AWS introduces a Managed Amazon Bedrock Knowledge Base that transforms retrieval into an agentic capability, replacing single embed-and-fetch calls with multi-turn planning and tool-driven retrieval. The AgenticRetrieveStream API enables a reasoning agent to decide whether to retrieve, which knowledge base to consult, and whether to iterate before composing a grounded, cited answer. This addresses limitations of ordinary RAG, such as answers spanning multiple corpora or requiring follow-up lookups after initial results. The service is now available as a first-class feature, eliminating the need to manually manage OpenSearch Serverless collections. AWS CDK users can now provision the knowledge base and AgenticRetrieveStream as stacks, though observability remains critical due to the non-deterministic paths agents take during retrieval.
Crypto
Anthropic files for IPO, targets November public offering
Anthropic filed an S-1 on Monday aiming for a November IPO, reporting first-half 2026 revenue of $16.3 billion and a net loss of $80.7 billion. The filing values the AI startup at $1.5–1.75 trillion, with investors expecting shares to price on Nov. 19 or 20. Chip stocks rose in premarket trading, and on Hyperliquid traders bet on the pre-listing share price using an ANTH contract that spiked to about $1,980 before falling 6% to $1,857.
Decrypt10.30Binance Loses $6.4 B in North‑Korean Hack, Users Covered by SAFU Fund
A North‑Korean hacker group stole roughly $6.4 B from Binance, the largest crypto theft ever, by compromising a DevOps engineer’s laptop. The stolen assets were spread across three blockchains: about $3.5 B in ETH, $1.4 B in BTC and $630 M on Tron. Stable‑coin issuers Tether and Circle used their built‑in freeze mechanisms to lock about $2.39 B, leaving $2.6‑$3.2 B still unaccounted for. Binance announced it had set aside $6.34 B from its SAFU fund to fully cover user losses and stated that user funds are “completely safe.” Aave also froze its V3 and V4 markets on Ethereum to prevent the attacker from using the stolen funds to create bad debt. The incident caused Bitcoin to dip below $77.5 k, ETH under $2,200, and BNB under $700, with over $1.33 B in positions liquidated.
Decrypt10.30TRM and Chainalysis Tie $6.4B Binance Hack to North Korea's TraderTraitor
Decrypt10.31Binance hacked for $6.4 billion in largest crypto heist ever, CEO says
Attackers stole roughly $6.4 billion from Binance on Thursday, the largest crypto hack in history, using a compromised DevOps engineer's laptop to access hot wallets across 10 blockchains. Binance says it will cover all losses with company funds and its emergency reserve, stating customer assets were never touched. The hack was attributed to North Korean group TraderTraitor, who used a self-destruct mechanism and EIP-7702 to drain assets.
Decrypt10.30Polymarket halts trading after security incident, users report unauthorized withdrawals
Polymarket paused trading and relayer actions after users reported unauthorized fund withdrawals. A wallet flagged as the suspected attacker holds about $2.65 million in POL after multiple USDC.e-to-POL swaps through Relay. Hudson Rock found a DebugBear-hosted script with wallet-draining code hard-coded to the same address. The platform's frontend may be compromised, with the trading API in major outage and the exchange in cancel-only mode.
Cointelegraph10.30North Korea’s TraderTraitor Group Linked to $6.3 B Binance Hack
Cybersecurity firms TRM Labs and Chainalysis, together with on‑chain investigator ZachXBT, have attributed the $6.3 B theft from Binance to the North Korean hacker collective TraderTraitor. TRM Labs cited both on‑chain and off‑chain intelligence to reach a “high confidence” attribution. ZachXBT noted that the stolen Binance funds and those taken in the earlier Bitget exploit were commingled at the same address after moving through THORChain, indicating a single actor behind both incidents. Chainalysis CEO Jonathan Levin estimated the loss at $6.4 B and linked it to TraderTraitor, citing similarities to previous North Korean‑linked hacks such as the Bybit and DMM Bitcoin thefts. The FBI has also identified TraderTraitor as the culprit in the $300 M DMM Bitcoin robbery, where an attacker used social engineering to gain employee access and divert funds. The attribution underscores a pattern of cross‑platform, state‑backed cyber‑criminal activity.
Binance Hit by $6.3 B Hack Linked to North Korea
Investigators have traced a $6.3 billion theft from Binance’s hot‑wallet signing infrastructure to a North Korean threat actor. Attackers exploited a vulnerability in Binance’s hot‑wallet signing system to authorize unauthorized transfers, then moved the stolen funds through a series of mixing services to obscure their trail. The incident remains under investigation, with authorities tracking the flow of the compromised assets.
Decrypt10.27SpaceX secures $40 billion GPU lease from Apollo, Musk warns of neocloud consolidation
SpaceX signed a binding commitment for approximately $40 billion in GPU leases from an Apollo-managed entity, with leases starting in the first half of 2027 and the master lease expected to close before the end of 2026. Elon Musk warned that debt markets have largely cut off funding for companies without substantial revenue outside AI data centers, predicting "guppy-level death and consolidation" among the 200+ neoclouds. SpaceX, which recently absorbed xAI, stated it is investment grade and has revenue from Starlink and rocket launches independent of AI.
Decrypt10.29SK Hynix Reports Record Q3 Revenue and Profit as AI Memory Demand Outpaces Supply
SK Hynix announced third‑quarter revenue of 104.6 trillion won (about $78 billion) and operating profit of 83.3 trillion won (about $65.5 billion), a 328 % and 632 % increase YoY respectively, giving the company an 80 % operating margin. It is the first time the South Korean memory‑chip maker’s quarterly sales exceeded 100 trillion won. The company said that demand for DRAM and high‑bandwidth memory (HBM) will continue to outstrip supply through 2030, and that its DRAM and HBM volumes through 2029 are effectively sold out. Long‑term agreements with 14 key customers cover more than half of SK Hynix’s 2027 and 2028 production capacity, locking in future demand. CFO Kim Woo‑hyun noted that the drop in net income from the second quarter was largely due to non‑operating expenses, including an 11.6 trillion won currency loss and a 22.9 trillion won valuation loss on investments.
Ukraine’s battlefield relies on Starlink as EU’s Iris² network remains years away
Ukraine’s military and civilians depend on Elon Musk’s Starlink satellite internet for real‑time communications and drone operations. A sudden slowdown last week exposed the vulnerability of relying on a single US‑based provider. Alternatives such as France’s Eutelsat/OneWeb constellation, with only about 600 active satellites, lag far behind Starlink’s 10,000‑plus network. The EU is developing a rival system, Iris², but it is not expected to be operational until 2029. Until then, Ukraine can only tap into the EU’s Govsatcom sharing network, which cannot support military‑grade tasks like drone control. Musk warned that cutting Starlink would collapse the Ukrainian front line, highlighting the strategic importance of satellite connectivity in modern warfare.
Finance & macro
Latent Space10.27Anthropic files S-1 for Nasdaq listing under ticker ANT, reports $16.3B H1 revenue and $974M Q2 operating profit
Anthropic filed its S-1 on Monday, planning a Nasdaq listing under ticker ANT. Revenue surged from $41 million in 2023 to $16.3 billion in H1 2026, with Q2 revenue reaching $11.6 billion, up 1,368% year-over-year. The company reported $974 million in Q2 operating profit and $3.9 billion in H1 operating cash flow, marking its first profitable quarter. The $80.7 billion H1 net loss is largely driven by $81.1 billion in non-cash charges tied to financial instruments.