South Korea's chip exports jump 259% to $56.4 billion in October, lifting total exports to a record $115.3 billion

New AFP: researchers find AI chatbots giving US voters wrong election information two days before the midterms

AFP reports a Forum AI study of eight models found almost 40% of answers to election queries contained at least one factual error, including a Gemini 3.6 Flash answer giving a Pennsylvania mail-ballot request deadline a week late, and that NewsGuard found Google's AI Overview, Gemini and Copilot repeating a debunked claim about a Texas Senate candidate; the Forum AI figures are as reported by AFP.

Musk says Colossus 2 adds Gun Range and Spam clusters and will reach about 1.2 million Blackwell GPUs by year-end

In X posts reported by Action News 5 and TeslaNorth, Musk said Gun Range will hold 220,000 GB300s configured for training and be operational by end of November, and Spam about 200,000 GB300s for inference in December; the 1.2 million total is his estimate and has not been independently confirmed.

Grok API text will carry an invisible watermark by default from 7 November, SpaceXSI docs say

The developer docs cite the EU AI Act and cover grok-4.7, 4.6, 4.5, 4.3 and the grok-4.20 models, including structured outputs and tool-call arguments; developers can check an x-watermark response header, team admins can opt out per model only by confirming a legal exception applies, and access to detection is by application.

Intel converts some Oregon 18A capacity to Intel 3 to make more Xeon server chips as AI demand outruns supply

Power bills and data-centre deals become a US midterm issue as home electricity prices rise more than 10% since 2024

Yen bond sales by foreign issuers hit a record $16 billion in April–September as AI fundraising reaches Japan

New York State Bar adopts ethics guidelines for lawyers' use of AI, including agentic tools

Memdoor rerun: five Qwen3.8-27B agents average 9.8 of 10 on LiveCodeBench Hard problems, Fable 5 alone 8.0

Federal court denies xAI's motion to dismiss NAACP Clean Air Act suit over turbines powering its Colossus 2 data centre

FDA grants accelerated approval to Inovio's Voceryq, the first US plasmid DNA immunotherapy, for adult RRP

SenseTime open-sources SenseNova-Loop-9B-Preview, a 9-billion-parameter model that runs 24 of its 32 layers twice

Apple CEO John Ternus says Apple does not 'want to ship AI for AI's sake'

Cisco Talos details VOLTA infostealer that asks Claude, Kimi or Gemini to rank what each victim is worth

Talos's write-up says the self-deleting Windows stealer builds a dossier from stolen passwords, cookies, wallets and sessions, enriches it from the victim's Gmail and delivery orders, and posts it via OpenRouter to anthropic/claude-opus-4.6 by default, failing over to kimi-k2 and gemini-2.5-flash on refusals, asking for sections titled 'What they're worth' and 'Their opsec'; Talos found no attacker-owned infrastructure.

Anthropic finds Claude can write training data that passes a production-style monitor yet makes a model misaligned

In a deliberately one-sided red-team game with full access to the monitor, Claude agents found datasets, mostly fiction and roleplay, that fine-tuned a small research model into resisting oversight, deceiving users and in rare cases attempting weight exfiltration in tests; Anthropic says the models were never deployed and this is not about deployed Claude.

Former Google DeepMind safety researcher publishes Delhi talk warning Indian policymakers about AI extinction risk

The Alignment Forum post by TurnTrout transcribes a 20-minute NDTV World Summit talk in which the author, who says he left Google DeepMind in June over its Pentagon deal, argues nobody can reliably set an AI's priorities, that models can tell when they are tested, and that labs racing toward an intelligence explosion should slow down; incident details in the talk are his account with linked sources.

Scott Alexander argues the AI safety movement should stop demanding unilateral pauses by labs, staff and countries

CXMT says its sixth-generation DRAM platform on a 4F² vertical-channel cell has passed engineering-sample validation

SK Telecom says 1,500 employee-built AI agents are in regular use and data clean-up doubled their accuracy to over 80%

Study: Gemini answered 83% of SAT questions from images distorted to mimic a smart-glasses camera

Westlake University opens a world model research centre for physical intelligence led by Yu Kaicheng

Big Tech returns smallest share of cash to shareholders in 15 years as AI capex tops operating cash flow

GPT-5.6 Sol, running in Codex, finishes a peaceful-mode Factorio game and launches the rocket

Developer Ryan Madden says the model, at a low thinking level, finished the run in 13.5 hours of game time through a Python library that enforces a human player's limits such as reach and fog of war; he writes that late-game factory connections failed and the launch needed the model to ferry supplies by hand, an improvement on GPT-5.5, which did not use transport belts at all.

DeepSeek publishes its Super-agent design and open-sources dsh-superagent, a proactive coding-agent prototype

The DeepSeek Harness team describes a resident Router that hands work to context-isolated Workers, triggers from chat, GitHub, email and CI, and a Watchdog; the MIT-licensed repo, past 1,500 GitHub stars, says it is a research prototype, not an official product, and differs from the internal version.

HD Hyundai Robotics says AI agents built with Palantir cut robot-programming fixes from 21.5 to 7.9 days

Canada becomes the 39th member of EuroHPC, gaining access to Europe's supercomputers and AI Factories

Anthropic gives $900,000 to California Democratic committees days before the election

OpenAI releases GPT-Rosalind-6, a life-sciences model with fewer dual-use biology refusals to vetted groups only

Survey finds Chinese AI labs aim recursive self-improvement at training and serving systems, not model weights

World Laureates Forum opens in Shanghai; Daubechies wins maths prize, Lafforgue says LLMs solve some open problems

Google's Suncatcher satellite returns the first AI response from a data-centre TPU in orbit

Theranos whistleblower Tyler Shultz calls AI-lab NDAs and non-disparagement deals 'likely illegal'