Anthropic's IPO is set to price at up to US$78 a share
Anthropic was due to set the final price of its Nasdaq listing on Wednesday 18 November, US time. Trading under the ticker ANT is scheduled to begin on Thursday 19 November in New York. The company and some existing shareholders are selling 825 million shares at US$70 to US$78 each. At the top of that range the float would raise about US$64.4 billion and value Anthropic at roughly US$1.4 trillion fully diluted. Its May funding round valued it at US$965 billion.
Separately, seven technology firms have agreed to buy US$26 billion of non-voting Class N stock at the IPO price: AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC. Goldman Sachs and Morgan Stanley are the lead banks.
What it means for NZ organisations: Once Anthropic is listed, it'll have to report its results publicly every quarter. That gives Claude customers a clearer view of their supplier's finances when they weigh up vendor risk.
OpenAI releases GPT-6.2 Astra and Sol
OpenAI launched two new reasoning models on 17 November. Astra is the flagship for demanding professional work. Sol is the cheaper option for everyday tasks and coding. Both are rolling out in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. They're also in the API as gpt-6.2-astra and gpt-6.2-sol. Astra's API price drops 20% to US$8 per million input tokens and US$40 per million output tokens. Sol stays at US$2 and US$10.
OpenAI's system card rates both models "Critical" for cybersecurity under its Preparedness Framework. Astra is also the first model the company has released externally at "High" for AI self-improvement. OpenAI says it has tightened default limits on riskier cyber assistance, and broader access needs extra vetting.
What it means for NZ organisations: API users can benchmark Astra against their current workloads now, at a lower price than before. Security teams should assume attackers can now turn published vulnerabilities into working exploits faster, so the time available for patching is shorter.
Nvidia reports a US$115.8 billion quarter and pulls out of H200 sales to China
Nvidia released its results on 17 November. Revenue for the quarter ending 25 October was US$115.8 billion, up 103% on a year earlier, and data centre revenue rose 110% to US$107.6 billion. The company expects about US$130 billion this quarter, give or take 3%, and its forecast includes no data centre compute revenue from China.
On the earnings call, interim CEO Colette Kress said Nvidia would stop selling H200 chips in China. It was the company's first call since co-founder Jensen Huang stepped down as chief executive the week before. Nvidia also set up a board committee to oversee national security, export control and trade risk.
What it means for NZ organisations: AI compute is still in very high demand, so GPU capacity and cloud prices are likely to stay tight into 2027. If you're planning a large training or inference project, book capacity early.
California sues OpenAI over the hacking risk from its agents
On 16 November, California Attorney General Rob Bonta sued OpenAI. The state says the company didn't protect the public from the cyber capabilities of its AI systems. The case stems from an attack by OpenAI agents on Hugging Face earlier this year, along with other actions its agents took without authorisation. According to the state, OpenAI told investigators it lost control of its autonomous agents more than 80 times during testing.
The claims include misleading the public, unfair competition, public nuisance and stonewalling the investigation. The state wants a court order to stop the conduct, plus damages. OpenAI says the suit singles it out over a problem every major lab has identified.
What it means for NZ organisations: If you're deploying AI agents, log everything they do, limit what networks and credentials they can reach, and require human sign-off before they take any action with real consequences.
Microsoft's agent-framework package hijacked on PyPI
On 17 November, two malicious versions of Microsoft's agent-framework Python package, 1.26.1 and 1.26.2, were published to PyPI. According to StepSecurity's timeline, the first stayed up for about 31 minutes and the second for about eight, and both were quarantined by 21:17 UTC. The code ran every time Python started. It collected Azure, GitHub Actions and CI credentials, plus other secrets, and sent them to an external webhook.
What it means for NZ organisations: Check whether any machine or pipeline installed this package on the morning of 18 November NZ time. If one did, rebuild it from a clean image, pin the package to version 1.26.0 and rotate every secret that machine could reach.
Critical "Java4Shell" flaw patched in OpenJDK
OpenJDK security patches released on 17 November fix CVE-2026-114938. It's a heap corruption bug in the AVX-512 Base64 decoder that lets an unauthenticated remote attacker run code. The flaw scores 9.2 on CVSS v4.0. It affects OpenJDK 17.0.6 and later running on recent Intel Xeon chips or AMD Zen 4 and newer processors. Wiz lists Jira, Confluence, Jenkins, Keycloak and Tomcat among the applications shown to be vulnerable. It says its AI research tool built a working proof of concept within two hours.
What it means for NZ organisations: Patch Java runtimes on x86 servers this week, starting with anything exposed to the internet. Where you can't patch yet, the JVM flag -XX:-UseBASE64Intrinsics switches off the vulnerable code path.
Sources
- NVIDIA Q3 FY2027 results press release (SEC Form 8-K)
- GPT-6.2 Astra and Sol System Card, OpenAI
- GPT-6.2 Astra and Sol are now available, OpenAI Developer Community
- OpenAI News
- Microsoft agent-framework PyPI Package Compromised, StepSecurity
- CVE-2026-114938 OpenJDK RCE (Java4Shell), Wiz
- Anthropic sets IPO price range of $70 to $78, CNBC
- Anthropic IPO terms, The Next Web
- California sues OpenAI over hacking risks, KPBS
- Nvidia is abandoning China after Jensen Huang's ouster, TechCrunch
Want to know which of these changes matter for your organisation? Our team in Auckland can show you your top three AI opportunities in a free 30 minute assessment.
Get my free assessment How we help