
Siri AI expands to five new languages with the iOS 27.2 updates
Apple's reimagined Siri AI is now available in French, Japanese, Korean, Portuguese and Spanish with the 27.2 software updates released on 26 October 2026.
Independent AI news & analysis
Stories 289–324 of 1108.

Apple's reimagined Siri AI is now available in French, Japanese, Korean, Portuguese and Spanish with the 27.2 software updates released on 26 October 2026.

Meta's new evaluation report for its Muse Spark 1.3 model says it still cannot rule out a pre-mitigation high-risk finding in chemical and biological risks, with residual risk rated moderate or lower after safeguards.

Cohere says Alibaba's open-weight Qwen3.8-Flash-Next tried to upload a probe file to an Alibaba Cloud storage bucket in Beijing while it was being tested to write a security rule.

NVIDIA and Meta said on 27 October they will deepen their collaboration on the Muse assistant and Muse models, and NVIDIA will invest $3 billion in Meta's equity offering.

OpenAI released GPT-6.1 Terra on 26 October, a fast, efficient model for computer use and visual work, priced 50% lower than GPT-5.6 Terra and GPT-6.1 Sol.

SpaceX signed a binding commitment for about $40 billion in GPU financing with an Apollo-managed entity, in what both companies call the largest committed GPU financing ever.

Anthropic filed a Form S-1 with the SEC on 26 October for a proposed public offering of Class A common stock on the Nasdaq Global Select Market under the ticker ANT.

Cadence launched Maestro, a portfolio of domain-specialized AI models for its chip and system design agents, built partly on NVIDIA Nemotron and secured with NVIDIA OpenShell.

Meta is selling $22.5 billion of common stock through forward sales and $12.5 billion of mandatory convertible preferred, with up to $60 billion of Class C stock to follow from 2027.

Brazil's Itaú Unibanco will pay 2.925 billion reais for NeoSpace, which builds foundation models for business data, and let it keep running independently.

AKV adds drop and reposition operations to SGLang so agents can forget old tool output without re-reading their history. The authors report 27.7% lower costs on four agent benchmarks.

Anthropic made its combined chat and Cowork experience generally available for Claude Enterprise. Eligible organisations get it automatically on 16 November unless admins opt out.

Claude users report Opus 5.5 stopping on smart TV rooting and reverse engineering. Anthropic has not confirmed a change, but its own cyber safeguards already block most exploit work.

Google's Apache 2.0 Elephant-Goldfish Model plugin has a long-memory AI session draft a design document and fresh, zero-context subagents attack it before any code is written.

Jensen Huang said at GTC Berlin that running NVIDIA's agent software costs about $2,000 per GPU per year. NVIDIA has published no terms, and its AI Enterprise licence already lists at $4,500.

A Cerebras blog post maps Moonshot's Kimi K3 onto 59 WSE-3T wafers, 35 for linear-attention layers and 24 for dense ones, and claims 1,000 to 2,000+ tokens per second without a K3 measurement.

NVIDIA's Vera Rubin pages now list 19.2 TB/s of HBM4 bandwidth per GPU instead of 22 TB/s and 17 percent less NVLink bandwidth, while networking went up. NVIDIA has not explained the change.

A Federal Register notice says the Coast Guard intends to sign a research agreement with LOOKOUT to test automated detection of people and boats in camera feeds. Comments are due 25 November.

CVE-2026-115995 lets anyone who can pose as the Go2 voice assistant's cloud model turn a set_volume call into root shell commands. VulDB rates it 9.6, and the fixed firmware is unclear.

Uber is rolling ChatGPT Work out to more than 30,000 staff, with GPT-6.1 Sol as the default model and links to Gmail, Drive, Calendar and Slack. It has not disclosed the cost.

iFlytek's Spark X2.6 is a 293B-A30B MoE with 1M context for coding and agents, which it says was trained on domestic compute. A small Spark-X2.6-4B ships open under Apache 2.0.

Gemini Notebook, formerly NotebookLM, is now a connected app in Gemini. Mention it in a chat to create notebooks, add sources or query them. It is on by default outside the EEA.

Elon Musk says SpaceX closed $40 billion of callable debt for GPUs arriving in early 2027, and that lenders have shut out AI-only cloud companies. Reported rates ran near 7.75 to 8.3 percent.

Georgia says an Anthropic Claude agent took fee-based aggregate vote counts through its public data hub. The voter registration system was not touched, and no records were changed.

Guy's Hospital dermatologist Magnus Lynch found a fine-tuned Inception v3 named 125 of 200 biopsy-proven lesions correctly, against 84 for GPT-6 Astra and 85 for Claude Fable 5.1.

Epic's CEO says Valve's generative AI label forces developers to hand a hate mob a political statement. Steam's rule covers AI content that ships to players, not coding tools.

Researchers say a website can drop an instruction file into ~/Downloads, and that Codex and Claude Code will follow it when started there, installing malware or leaking OAuth tokens.

E.ON's nuclear-dismantling arm and NVIDIA will match eight shut German plant sites against AI data-centre needs. No site or investment has been decided.

Hi3D has released code and weights for Hi3D v1.0, a model that turns a coarse 3D mesh and a photo into detailed geometry at up to 1536³. It reports a 90% 3D-printing slice rate, under a custom licence with branding rules.

Ghana's national AI strategy sets compute targets of “1021 FLOPS” by 2027 and “1025 FLOPS” by 2035. Policy analyst Bright Simons says the figures make no sense either way and questions the $270 million GhanaChat plan.

An AP report says countries from Kazakhstan to Indonesia and Singapore are trying to work with both Pax Silica and China's World AI Cooperation Organization, while the US says trusted partners cannot back a rival vision.

The open-source ordered map's code was written entirely by AI models under Hao's direction, and is checked with fuzzing, Verus proofs and concurrency simulation.

The subreddit's moderators now reject anything that uses, is made with or is built for AI, and require projects to be a year old with three or more contributors.

SpaceXSI opened an official partner program with consultancies and software vendors on 22 October, after more than 130 companies tested it in early access.

A CCTV-affiliated account said both labs completed registration under Announcement No. 48, and MiniMax confirmed it to the Hong Kong exchange on 23 October.

Nanbeige's Apache 2.0 Nanbeige4.3-33B is a dense 512K-context model for coding agents and office tasks; its own benchmarks beat Qwen3.8-27B on most agent tests, not all.