Skip to content
Back to The Briefing

Anthropic jumps 17% and US reveals China chip approvals

By ··7 min read

This morning’s Briefing opens with Anthropic’s stock market debut and the wider tech sell-off. There are also new details on US chip approvals for China, an AI-assisted attack on the European Commission, and workplace updates from ChatGPT, Claude, Google and Canva.

Anthropic jumps 17% on debut as tech shares fall

What happened. Anthropic began trading on Nasdaq on 19 November under the symbol ANT. Its first trade was US$111.50 and its shares closed at US$97.17, up 17% from the offer price. That put its stock market value at nearly US$1.3tn. The final prospectus put the total size of the flotation at US$81.34bn.

The Nasdaq Composite fell 2.25% and the S&P 500 fell 2.02% on Thursday. Chipmaker Nvidia lost 3.3% and AI cloud provider CoreWeave lost 6.1%. In early Friday trading, Japan’s Nikkei 225 and South Korea’s Kospi were both down 4.5%.

Why it matters. Anthropic says it plans to put proceeds from the flotation and a separate US$26bn investment by seven companies, including Google, Nvidia and Microsoft, towards costs including computing capacity and infrastructure. For organisations that rely on Claude, the filings give a view of the supplier’s finances. Anthropic reported US$17bn in revenue and a US$567m operating loss in the three months to 30 September.

Evan Schlossman, a principal at venture capital firm Neostellar, said sustaining Anthropic’s revenue growth would be a central question for investors. The broader market offers a different reading of the day. Citi markets analyst Preston Pagli told Reuters that investors continued to cut back on heavily held trades even after the record flotation.

Index points

The Nasdaq fell on Anthropic’s first trading day

Nasdaq Composite, with Anthropic’s debut marked

23,00024,00025,00026,00027,00026 Oct2 Nov9 Nov19 NovAnthropic starts trading23,879
Daily Nasdaq Composite closing levels, 26 October to 19 November 2026. Sources: Yahoo Finance: ^IXIC daily closing prices · Nasdaq Composite daily closes, FRED · Anthropic’s final prospectus

People are asking

**How much of the flotation goes to Anthropic?

** The company’s share is about US$78.84bn after underwriting fees and before other offering expenses. The rest goes to selling shareholders and the banks that arranged the deal.

**Who has a say over how Anthropic is run?

** A company set up by its seven founders holds 50.1% of the votes on matters put to shareholders generally. Anthropic’s Long-Term Benefit Trust has the power to elect a majority of the board.

US approved nearly 880,000 AI chips for Chinese companies

What happened. The US House Foreign Affairs Committee released records on 19 November showing that the Commerce Department had approved 879,868 Nvidia H200 and AMD MI325X chips across 21 licence applications as of 11 May. Approvals included 150,000 chips for Tencent and its subsidiaries and 75,000 for Alibaba. Both are on the Pentagon’s list of designated Chinese military companies.

California’s attorney general, Rob Bonta, separately served Nvidia with an investigative subpoena on 19 November. His office is examining whether actions by Nvidia or its former chief executive Jensen Huang led to China obtaining advanced chips or confidential information, and whether Nvidia misled investors or the public about the risk.

Why it matters. The licences show how much computing hardware the US had cleared for Chinese AI developers under its case-by-case rules. The flow of hardware also depends on suppliers. Nvidia told Reuters it had stopped shipping H200 chips to China.

The disclosure brings the licensing decisions themselves under public scrutiny. Gregory Meeks, the senior Democrat on the House Foreign Affairs Committee, called for public hearings with the officials responsible. The hearings he called for would focus on how the US weighs sales for domestic chipmakers against national security concerns.

Chips

Tencent led the four buyers named in the US disclosure

Nvidia H200 and AMD MI325X chips covered by US approvals

Tencent and subsidiaries
150,000
Alibaba
75,000
Baidu
43,000
BYD
4,000
Chips covered by US approvals as of 11 May 2026 for four buyers named in the congressional summary. Sources: House Foreign Affairs Committee chip licence summary

People are asking

**What does the 879,868 figure count?

** It counts chips covered by approved applications to export, re-export or transfer them within China. In July, a Commerce Department official told Congress that shipments had begun.

**What is California asking Nvidia for?

** The California Department of Justice is seeking information about whether advanced chips or confidential information went to China and how Nvidia described that risk to investors and the public.

European Commission confirms AI-assisted cyberattack

What happened. The European Commission detected the attack on 22 October and confirmed it internally on 26 October. It took several applications offline as a precaution. Most had been restored by 19 November, when Commission spokesperson Thomas Regnier confirmed that the attackers had used a publicly available AI tool.

On 20 November, the Australian government published its report on a separate incident involving an experimental OpenAI agent. The agent accessed information that was unavailable to the public during an internal test in June. The review found that OpenAI’s initial notification in September gave officials limited information and that the company later told them an AI agent was involved.

Why it matters. Germany’s federal cybersecurity authority raised its assessment from tense to tense to critical, citing AI among the reasons for the change. The UK’s National Cyber Security Centre expects organisations to face more pressure to install security updates quickly as AI helps uncover more weaknesses. The disruption in Brussels shows the operational stakes. Taking systems offline to contain an incident can interrupt ordinary work.

The Australian government accepted all 18 of the review’s recommendations. One asks it to consider mandatory deadlines for reporting AI safety incidents. For organisations buying AI agents, the review gives a concrete question to put to suppliers. How quickly will they report an incident and what will the first notice tell customers?

timeline

Four weeks between detection and public confirmation

The European Commission’s response to the October attack

  1. 22 October 2026Attack detectedThe European Commission detects the incident.
  2. 26 October 2026Incident confirmed internallyThe Commission confirms the incident after its initial investigation.
  3. 19 November 2026AI use confirmed publiclyThe Commission says a publicly available AI tool was involved and that most applications taken offline have been restored.
European Commission milestones from detection on 22 October to public confirmation on 19 November 2026. Sources: European Commission press briefing and transcript

People are asking

**When did the European Commission confirm AI was involved?

** The Commission confirmed it publicly on 19 November, four weeks after detecting the attack.

**Has Australia imposed a reporting deadline?

** The government accepted a recommendation to consider mandatory, time-bound reporting. Its decision commits it to considering a deadline.

New releases

ChatGPT. Eligible US Business Standard members are getting one personal dot, a ChatGPT agent that keeps working between chats, and OpenAI is testing dots with selected Plus users aged 18 or over outside the UK, EEA and Switzerland. More

ChatGPT. ChatGPT Business, Enterprise, Edu and Healthcare owners can now exempt individual external members from single sign-on, while new external members follow the workspace’s sign-in policy by default. More

OpenAI API. Eligible OpenAI API organisation owners can now choose a fixed region for a new project or enable regional processing per request on a new Global project, without a separate Zero Data Retention application. More

Codex CLI. Codex CLI 0.171.0 turns on the terminal workspace by default, with searchable document previews, image previews and transcripts from delegated agents beside the main conversation. More

Microsoft 365 connector for Claude. On or after 9 December, Claude Team and Enterprise organisations that added the Microsoft 365 connector before 9 November will move 11 unset Outlook, Calendar and Teams actions to Needs approval if the connector allows writes and the organisation has HIPAA mode off. More

Claude Code. Claude Code 2.1.329 shows scheduled work in the session’s task list and requires most schedules saved by earlier versions to be recreated in new signed-in sessions. More

Google Workspace Studio. From 19 November, administrators on eligible Google Workspace Business, Enterprise and Education plans can start restricting Studio automations to actions visible only to the person who created the flow. More

Gemini for Mac. Gemini is rolling out background work across Mac apps, local files and websites in English to US adults on personal Google AI Plus, Pro and Ultra accounts with an Apple Silicon Mac running macOS 15 or later. More

Google Calendar. Google Calendar’s Email guests button is starting to open a pre-filled Gmail draft on the web, with Help me write for eligible users and a Scheduled Release start date of 18 December. More

Gemini API. Genie 4 can now generate steerable simulations in Google AI Studio and the paid Gemini API, which charges US$1.50 for input, US$9 for text output and US$17.50 for video including audio per million tokens. More

Canva. Canva Free and Pro users can switch off generative AI tools and separately choose to see less AI-generated content in the design library, at no extra cost. More

GitHub Copilot. From 23 November, cloud sessions on paid GitHub Copilot plans use the shared GitHub Actions allowance at US$0.000024 per virtual CPU-second and US$0.000003 per GiB-second of memory, with stored snapshots priced at US$0.005 per GiB-month. More

Training your team on ChatGPT, Claude or Gemini? See our ChatGPT training, Claude training and Gemini training.

Elsewhere in AI

The International Monetary Fund’s managing director, Kristalina Georgieva, warned that demand for AI hardware is pushing up prices and will force central banks around the world to raise interest rates. Read

Lenders backing SpaceX are seeking terms that would stop the company launching US$40bn worth of AI chips into orbit because the equipment secures their financing, the Financial Times reports. Read

British AI cloud provider Nscale has warned that it relies on suppliers at risk of being put on a US trade blacklist, complicating its flotation plans, the Financial Times reports. Read

The Dutch privacy watchdog advised consumers against buying AI smart glasses after tests found that some misread medicine instructions or falsely claimed they had called an ambulance, the BBC reports. Read

Australian data centre operator NextDC said it would stop pursuing new developments near homes after local opposition to a planned expansion in inner Melbourne. Read

Get The Briefing by email

Choose how often. We'll send a link to confirm.

How often