Daily Summary of AI · Live edition, updated continuouslyWednesday 18 November 2026 · UTC

Latest AI News Today: 18 November 2026

Anthropic's IPO nears as Huang faces Senate probe and Nvidia revenue doubles

Anthropic is preparing a blockbuster Nasdaq IPO that could value the Claude maker at $1.405 trillion, while former Nvidia CEO Jensen Huang faces Senate testimony over his affair with a suspected Chinese spy. Nvidia reported quarterly revenue of $115.84 billion, more than doubling, as its interim CEO said AI demand remains strong. OpenAI also said its models delivered a year's worth of chip optimization work with AMD.

Illustration of the day's news: Anthropic's IPO nears as Huang faces Senate probe and Nvidia revenue doublesIllustration
DS
Daily Summary Of Editors · How we summarise
19 stories23 sources
Image from Illustration: OpenAI opens AI systems to four independent onsite auditors for the first timeIllustration
01First reported

OpenAI opens AI systems to four independent onsite auditors for the first time

OpenAI is giving Apollo Research, METR, Transluce and Goodfire access to unreleased models, source code, training data and monitoring systems to check safety and model behavior. An independent safety board oversees the process, receives unredacted reports and mediates disputes, while auditors publish findings. OpenAI can redact confidential material and delay reports to fix risks, fulfilling White House Accord on Super Intelligence commitments.

Developing storyOpenAI GPT-6 Astra safety monitoring reportsFull timeline →
Image from Illustration: Jensen Huang and Lutnick agree to appear before Senate Banking CommitteeIllustration
02First reported

Jensen Huang and Lutnick agree to appear before Senate Banking Committee

Former Nvidia CEO Jensen Huang has agreed to appear before the Senate Banking Committee, and Commerce Secretary Howard Lutnick has also agreed to testify, Chair Tim Scott said. The voluntary appearances follow pressure over Huang's affair with a suspected Chinese spy and Nvidia's chip sales to Megaspeed. Huang also agreed to a closed-door interview before three House committees on Dec. 1.

Developing storyJensen Huang resigns over spy allegationsFull timeline →
Image from Illustration: Humyn Labs raises $47 million to train robots with human dataIllustration
03First reported

Humyn Labs raises $47 million to train robots with human data

Humyn Labs, based in Bengaluru, raised $47 million in Series A financing, including $26 million in equity led by Accel and $21 million in venture debt led by Lighthouse Canton. The company records people performing everyday tasks with camera hardware, then processes and labels the footage to train AI and robotics models. It says it has 12 paying customers, with about 90% of revenue from the U.S.

Image from Illustration: China's military seeks Nvidia's advanced chips, report saysIllustration
04First reported

China's military seeks Nvidia's advanced chips, report says

A congressional commission says China's military is actively seeking Nvidia's most advanced chips for AI capabilities, while U.S. export controls aim to limit access. The report says Chinese firms have used remote access to Nvidia chips in third-country data centers, including Malaysia, and that a substantial compute gap remains. The findings come after Nvidia's CEO Jensen Huang resigned amid reports of an affair with a suspected Chinese spy.

Image from Illustration: Suno adds Visuals to pair AI music with generated videoIllustration
05First reported

Suno adds Visuals to pair AI music with generated video

Suno has launched Visuals, a feature that turns its AI songs into video clips of up to 15 seconds or images, with lip sync, lyric overlays and editing tools. The feature runs on H3 Move, a customized MiniMax model, and uses separate Visuals credits. Filters and an embedded watermark are included as Suno faces rulings and lawsuits from music rights groups and labels.

Image from Illustration: SID AI releases SID-2 agentic search model and SID-2 SwarmIllustration
06First reported

SID AI releases SID-2 agentic search model and SID-2 Swarm

SID AI has released SID-2, an agentic search model that searches data, reads results and returns ranked document IDs, plus SID-2 Swarm, which runs multiple agents over separate corpus slices. The company says it replaces fixed RAG pipelines by deciding what to search next. On its self-reported 506-question benchmark, high-effort SID-2 scores 0.14 to 0.22 higher per metric than the base model.

Image from Illustration: South Korea offers $75,000 rewards and emergency alerts to stop exam cheatingIllustration
07First reported

South Korea offers $75,000 rewards and emergency alerts to stop exam cheating

South Korea will send emergency alerts to residents' phones before its Suneung exam and offer at least $75,000 for reporting cheaters, President Lee Jae Myung said. Officials will inspect glasses, cameras and earbuds, deploy about 75,000 supervisors and 8,400 additional personnel, and seek records from AI companies of users photographing exam questions. The measures aim to protect a key exam expected to draw about 550,000 students.

Image from Illustration: Anthropic adds Code Mode to Claude Code for chained tool callsIllustration
08First reported

Anthropic adds Code Mode to Claude Code for chained tool calls

Anthropic has added Code Mode to Claude Code as an experimental, opt-in feature that lets Claude write JavaScript to run several tool calls in one step. The mode, shipped in Claude Code v2.1.323, can run commands in parallel, pass results between calls and return only the output Claude needs. Anthropic said an example using code execution with MCP cut token use from 150,000 to 2,000.

Image from Illustration: Over a third of Gen Z rely on AI for friendshipIllustration
09First reported

Over a third of Gen Z rely on AI for friendship

A Gallup and Boston Children's Hospital Digital Wellness Lab report found 33 percent of Gen Z rely on AI for companionship, rising to 50 percent among daily AI users and 42 percent among those with online-only friends. The survey of 15,925 people aged 15 to 22 in 16 countries found in-person friendships were rated highest for strength and support. Researchers said reliance on online-only friends or AI may raise the odds of loneliness.

Image from Illustration: Cambridge dictionary names 'human' word of the yearIllustration
10First reported

Cambridge dictionary names 'human' word of the year

Cambridge dictionary named human its word of the year for 2026, defining it as having the qualities, faults and feelings that people have, as opposed to gods, animals or machines. Editor Colin McIntosh said AI is changing how people talk about being human and has led to the word distinguishing human work from advanced computer systems. The choice reflects debate over what humans may lose when they outsource thinking to AI.

Image from Illustration: AI finds critical Java flaw called Java4ShellIllustration
11First reported

AI finds critical Java flaw called Java4Shell

Calif used AI to find a critical Java memory flaw called Java4Shell, which could let a remote attacker run unwanted commands. Java is used by top companies, banks, hospitals and governments, so the bug is a concern even though Oracle has not acknowledged it. Oracle released a critical patch update on Tuesday, and OpenAI's not yet public GPT-6 Cyber model was used in the discovery.

Developing storyCritical Java4Shell flaw in OpenJDK Base64 decoderFull timeline →
Image from Illustration: Nvidia revenue doubles as Kress says AI demand remains strongIllustration
12First reported

Nvidia revenue doubles as Kress says AI demand remains strong

Nvidia reported quarterly revenue of $115.84 billion, more than doubling, and projected revenue of about $130 billion for the current period, roughly 90% growth. Interim CEO Colette Kress said customers, suppliers and partners remain fully supportive and raised next year's growth forecast from 70% to 75%. Shares rose almost 5% in extended trading after the report.

Developing storyNvidia tops $100 billion in a quarterFull timeline →
Image from Illustration: OpenAI bugging case defendant Alfozan set for January trialIllustration
13First reported

OpenAI bugging case defendant Alfozan set for January trial

A federal judge set Jan. 20 for the trial of Abdulrahman Alfozan, a former OpenAI engineering lead accused of planting a recording device in a San Francisco conference room. Alfozan, who pleaded not guilty to one count of intercepting oral communications, faces up to five years. His lawyer sought a speedy trial after prosecutors were slow to provide evidence, and the judge ordered FBI reports by Nov. 24.

Image from Illustration: Nvidia's Kress says customers and employees are not going anywhereIllustration
14First reported

Nvidia's Kress says customers and employees are not going anywhere

Interim CEO Colette Kress told CNBC that Nvidia's customers and employees remain committed after Jensen Huang's abrupt departure last week. She said hyperscalers are seeking more compute and that customers have contacted her saying they need more demand. Kress also said the leadership team and workforce support are intact, countering concerns that Huang's exit could push customers to in house chips or competitors.

Developing storyJensen Huang resigns over spy allegationsFull timeline →
Image from Illustration: Microsoft makes Word, Excel and PowerPoint generally available inside CopilotIllustration
15First reported

Microsoft makes Word, Excel and PowerPoint generally available inside Copilot

Microsoft said Word, Excel and PowerPoint are now generally available inside Copilot, which CEO Satya Nadella called the new OS for work. Users can create documents, spreadsheets and presentations in Copilot, with files remaining editable in the Office apps. Copilot Autopilot, a proactive agent, was also announced, and Microsoft IQ became generally available, with Work IQ giving agents context from emails, meetings, chats and files.

Image from Illustration: Ofqual urges life altering penalties for students using AI or smart glassesIllustration
16First reported

Ofqual urges life altering penalties for students using AI or smart glasses

Ofqual head Ian Bauckham said students could be disqualified from GCSEs, A levels and other qualifications if caught using AI or smart glasses to cheat. He called for retailers to be barred from marketing wearable technology for exam cheating and said invigilators should be trained to spot emerging risks. Ofqual recorded 2,225 cases of smart device and mobile phone cheating during GCSE, AS and A levels in summer 2025.

Image from Illustration: EU commits 400 million euros to PIX Europe photonic chip projectIllustration
17First reported

EU commits 400 million euros to PIX Europe photonic chip project

The European Union committed 400 million euros to PIX Europe, a project aimed at making the continent a global leader in photonic integrated chips. Photonic chips use light instead of electricity, and are described as faster, using less energy and generating less heat than traditional computer chips. A Dutch team is building a 10,000 square metre factory in Eindhoven to produce six inch indium phosphide wafers, due to fully open in 2030.

Image from Illustration: Reson8 open sources Rumble-2-Flash speech model for 40+ languagesIllustration
18First reported

Reson8 open sources Rumble-2-Flash speech model for 40+ languages

Reson8 has open sourced Rumble-2-Flash under the Apache 2.0 license. The 0.5 billion parameter speech recognition model supports more than 40 languages and is aimed at developers needing fast, non-English transcription. The company says it averages a 6.96 percent word error rate on the Open ASR Leaderboard multilingual track, while SiliconANGLE cites the company's own runs at 7.09 percent across seven languages.

Image from Illustration: OpenAI says its models delivered a year's worth of AMD chip optimizationIllustration
19First reported

OpenAI says its models delivered a year's worth of AMD chip optimization

OpenAI says its models achieved up to 14 percent power, performance and chip area improvement on AMD's flagship GPU IP in under five months, which it calls about a year's worth of progress beyond AMD's prior optimization. AMD has folded some changes into a design sent to manufacturing, OpenAI says. The company also used its Astra model to improve RISC-V core designs and Jalapeño silicon, and is working with Synopsys on GPT-Synopsys.