What you need to know
5 storiesThe short versionAI tools helped find critical flaws in Java and nginx, and Newsom's office proposed giving California the power to shut down a dangerous AI model.Full rundown
If you run Java or nginx on a server, patch it. Calif says its researcher Quang Luong found Java4Shell, a memory bug in OpenJDK's Base64 decoder, with OpenAI's Codex in about an hour, and Oracle shipped fixed builds on November 17. Xint disclosed five nginx flaws that each let an attacker run code without logging in, and F5 fixed them the same day in nginx 1.30.7 and 1.31.9. An automated Xint scan found four of them.
OpenAI released GPT-6.2 Astra and Sol. Astra costs 20% less than GPT-6 Astra in the API and ties Claude Fable 5.2 at 174 on Epoch's capabilities index, but on tasks drawn from Epoch's own research work it fell behind Fable 5.2. Artificial Analysis measured a 59% hallucination rate, so check what it writes.
Governor Gavin Newsom's office released proposals that would let California order a dangerous AI model shut down in an emergency and cap AI that builds its own successors. None of it is law yet. The Center for AI Safety published a statement that Anthropic is not a safe AI company as it heads into its IPO, and 92 people had signed it by that afternoon.
Nvidia reported record revenue of $115.8 billion in its first quarter since Huang left, and its outlook of about $130 billion assumes no data center compute sales in China. Run a LiteLLM proxy? Update it to 1.83.11, since CISA now lists its guardrail sandbox escape as exploited.
- 1
Calif says one researcher using OpenAI's Codex found a critical Java flaw in an hour
The AI cost under $200, and Calif says it got into Jira and Jenkins without a password.
- 2
Five teams independently found two critical nginx flaws, and Xint's automated scan found more
Wiz says five teams found the same nginx bugs on their own, and nginx's advisory list went from two in 2025 to 26 this year.
- 3
OpenAI's GPT-6.2 Astra ties Claude Fable 5.2 on Epoch's index, but trails it on real tasks
Astra is 20% cheaper and sets a math record, but Epoch says it left out estimates a contractor couldn't.
- 4
Newsom proposes state power to shut down dangerous AI models and cap self-improving AI
The plan would let the state cap AI that designs its own successors, and four advisers say the labs face less oversight than their cafeterias.

- 5
Steven Pinker and Beff Jezos sign CAIS's statement that Anthropic is not a safe AI company
AI safety researchers, PauseAI and e/acc's founder signed the same paragraph, and Dan Hendrycks says other groups were afraid to.

Running stories4

California AI executive order
- Nov 18Newsom proposes state power to shut down dangerous AI models and cap self-improving AI
- Sep 18Newsom gave experts two months to recommend an AI kill switch and auditors inside frontier labs
Photo: Christopher Michel, CC BY-SA 4.0Anthropic ipo
- Nov 18Steven Pinker and Beff Jezos sign CAIS's statement that Anthropic is not a safe AI company
- Nov 12Anthropic says Mythos got Claude running on AMD chips in three days, almost unaided
- Nov 10Anthropic starts its IPO roadshow, but a single founders' share will carry 50.1% of the vote
Photo: Dicklyon, CC BY-SA 4.0Nvidia huang megaspeed briefing
- Nov 18Nvidia's first results without Huang assume no China data center sales next quarter
- Nov 15Nathan Lambert says Jensen Huang's exit from Nvidia is a major setback for US open models
- Nov 14Musk backs a petition to reinstate Jensen Huang, and Gerstner and Baker write Nvidia's board
Photo: World Economic Forum, CC BY 3.0Embedded evaluators
- Nov 17OpenAI brings four outside evaluators inside, but one says the access isn't employee-like
- Oct 15Anthropic embeds six METR evaluators with employee-like access, and can't veto what they publish
- Sep 19Anthropic's first embedded evaluator is the consulting firm Accenture, but Anthropic is paying for the work
Also that day1
In brief10
Anthropicv2.1.328
Claude Code 2.1.328 brings auto mode to Opus 4.6 and Sonnet 4.6 on Amazon Bedrock, Google Cloud, Microsoft Foundry and the Claude apps gateway, and lets a failing MCP tool hook block the action. github.com →
GoogleIntroducing Gemini 4 Neon
Google DeepMind introduced Gemini 4 Neon, a new model in its Gemini 4 line, on its blog on November 18, 2026. deepmind.google →
GoogleGenie 4 is now available
Google DeepMind announced on its blog on November 18, 2026 that Genie 4 is now available. deepmind.google →
GoogleThe Gemini app is now available on lower-memory Android devices.
Google's Gemini app now runs on lower-memory Android phones, and those users get Gemini Live and custom image creation too. blog.google →
OpenAIInside the Codex sandbox
OpenAI explained how Codex fences off the commands it runs on your own computer, with network access switched off unless the sandbox settings allow it. openai.com →
5 more
OpenAIResearch acceleration: The jagged frontier
OpenAI published research on how coding agents are speeding up its AI research while their performance stays uneven, and on what that means for recursive self-improvement. openai.com →
OpenAIManaging frontier AI risk: From training to deployment
OpenAI laid out how it handles risk from training to deployment, saying it trains frontier models without live internet access and now monitors them across most internal and external deployments. openai.com →
OpenAILe Groupe Figaro and OpenAI partner to bring Le Figaro journalism to ChatGPT
Le Groupe Figaro signed a partnership with OpenAI to bring Le Figaro's journalism into ChatGPT, OpenAI announced on November 18, 2026. openai.com →
OpenAIAdvancing chip design with OpenAI models
OpenAI says its models, including GPT-6 Astra, optimized circuits used in Jalapeño silicon and improved RISC-V core designs. openai.com →
OpenAIThe Silent Consensus
OpenAI's Dean Ball published an essay rejecting what he calls the industry's silent consensus that AI will inevitably replace people as the force that runs society. openai.com →
From the labs' own feeds12
Releases 3
2.1.328Claude Code2 preCodex CLI3.55.1OpenAI Python SDKLab posts 1
Status 5
- Outage · OpenAICosts API data is delayed
- Outage · AnthropicClaude Meetings auto-join limited on the Team plan
- Outage · OpenAIElevated conversation errors for Free and Go users
- Deprecation · OpenAIOpenAI API deprecations: page updated (4 added, 0 removed)
- Deprecation · AnthropicAnthropic model deprecations: page updated (30 added, 26 removed)
Other days
Every dayRun AI at home
Can I run it?By topic
Anthropic
More
Photo: TechCrunch, CC BY 2.0Anthropic starts its IPO roadshow, but a single founders' share will carry 50.1% of the vote
The public gets one vote a share, the chipmakers buying $26 billion of stock get none, and a single share held by the founders outvotes them all.
November 10, 2026Anthropic cuts its bug disclosure deadline to 30 days, saying another AI will find the bug
Nov 16Anthropic says Mythos got Claude running on AMD chips in three days, almost unaided
Nov 12Anthropic launches Claude Sheets and Whiteboards, but Claude Meetings is still coming soon
Nov 10AI Futures Project now expects a superhuman AI researcher in 2027, citing METR's Anthropic report
Nov 10Anthropic releases Fable 5.2 and Mythos 5.2, but limited their skill at speeding up AI research
Nov 9
OpenAI
More
Photo: Jernej Furman from Slovenia, CC BY 2.0ChatGPT Wallet lets OpenAI's dots buy things for you, but you approve each purchase
Each purchase needs Face ID or Touch ID, but a subscription you approve keeps charging after that.
November 14, 2026AI safety
More
Photo: AgnosticPreachersKid, CC BY-SA 3.0Four nonprofits sue Anthropic, Google, Meta, OpenAI and SpaceXSI to freeze frontier AI development
The groups are using a California consumer law and the executives' own warnings to ask for a freeze.
November 13, 2026Ready-made prompts
All 59 promptsPick one, change the bits that are yours, paste it into ChatGPT or Claude.
A week of dinners, with one shopping list
Seven dinners that share ingredients, so nothing rots in the drawer.
CookingCatch me up before the new season
You watched it two years ago. Get the plot back in five minutes, no spoilers past where you stopped.
Stories & mangaChase money someone owes me
The message to send today, and what to do when they ignore it.
Writing & memosA study plan for the time I've actually got
Works backwards from the date, and tells you what to drop when you fall behind.
LearningA present for someone you actually know
Tell it what they're really like. Get ideas that aren't a gift guide for their demographic.
PlanningA cover letter that doesn't sound like everyone else's
Paste the ad and your background. Get a short letter that points at real things you've done.
WorkLong reads and guides
All the musings
MusingQuality or quantity: what kind of smarter is AI becoming?
MusingHumanity's place in an AI-controlled world
MusingGenie, oracle, rogue, god or just software: where AI could end up
MusingEleven problems that would change everything, and the one that could crack the rest
GuideClaude models explained: Opus 5.5, Fable 5.1, Mythos and why there's no Opus 5.1
Guide