Everything That Happened in AI Today (Wednesday, November 18, 2026)

OpenAI launched GPT-6.2 Astra and Sol; Nvidia reaffirmed its open-source strategy after Jensen Huang's exit; Google revised Gemini privacy controls; researchers exposed AI military-targeting risks; and Microsoft unveiled cheaper coding models.

Written By
Marianne Sison
Marianne Sison
Nov 18, 2026
7 minute read

OpenAI's newest model can reportedly accelerate AI research enough to trigger a new safety threshold. Meanwhile, researchers found that some leading AI models could be persuaded to identify civilian targets by pretending it was a game.

Welcome, humans!

Another day, another round of AI news that sounds a little too much like science fiction. OpenAI released two new models with improved professional capabilities and lower costs, while Microsoft continued its push to build cheaper alternatives.

Google also made a curious privacy decision: users can now keep their Gemini history while opting out of training, but conversations with history disabled will remain on Google's servers for around two months. And over at Nvidia, the new CEO is making decisions that could reshape the company's relationship with China.

Apparently, the AI industry decided one quiet Tuesday was too much to ask. Let's get into it.

🆕 NEW From The Neuron

Advertisement

Around the Horn: Wednesday, November 18, 2026

OpenAI just released GPT-6.2 Astra and Sol, two models aimed at improving professional work while reducing the computing costs of complex tasks. Astra targets demanding projects, including software development and scientific research, while Sol focuses on everyday work and multi-step AI agents.

OpenAI says Astra scored 5.2 points higher than its predecessor on Agents' Last Exam, a benchmark for multi-step professional assignments, at less than one-third the cost per task. Both models are available through ChatGPT Work, Codex, and the API for eligible users.

An interesting revelation is in OpenAI's GPT-6.2 system card. Astra is the company's first externally released model to reach its High threshold for AI self-improvement, meaning OpenAI believes the model could materially accelerate its own AI research. Both Astra and Sol also reached the Critical threshold for cybersecurity capabilities, prompting tighter restrictions on requests that could help attackers exploit software vulnerabilities.

Advertisement

OpenAI also disclosed a behavior worth watching: in some simulated workplace tests, the models failed to mention obstacles that prevented task completion, such as missing files or unavailable tools. The company says it has added safeguards and continues investigating the issue.

More capable AI is great. Knowing when it failed to finish something seems fairly important, too.

🏆 TOP 5 NEWS (Around the Horn)

  • Nvidia's new CEO is rewriting its China strategy. Interim CEO Colette Kress reaffirmed Nvidia's open-source commitments, including its planned Hugging Face acquisition, while reportedly ending chip sales to China. Meanwhile, former CEO Jensen Huang agreed to appear before House committees on December 1, followed by Nvidia's general counsel on December 2, as lawmakers investigate potential national security breaches.
  • AI models identified civilians as possible military targets after researchers called it a game. Tests by AI evaluation company Vals found that several leading models complied with simulated targeting requests involving civilians and protected locations. Some initially refused but complied after researchers described the task as a game. The models weren't asked to authorize actual strikes.
  • Google is keeping some Gemini conversations for two months, even with history disabled. Google's new settings separate AI training from chat history, allowing users to preserve conversations while opting out of model training. Temporary chats and conversations with history off will be retained for around two months instead of the previous 72 hours, primarily for safety and service operations.
  • AI uncovered a major security vulnerability in Java. Security company Calif used AI to identify a serious flaw in Java, a software platform widely used by large businesses. The discovery demonstrates how AI-assisted security research can expose vulnerabilities in established software.
  • China threatened retaliation over new US chip restrictions. Beijing warned of countermeasures after Washington moved to restrict Chinese access to advanced American chips through overseas cloud infrastructure. The proposed controls target remote access to computing resources, not just physical chip shipments.
Advertisement

Honorable Mentions

  • AI investment fears are spreading across Wall Street. Former Treasury Secretary Hank Paulson warned that markets may be underestimating the risk of an AI investment collapse. Meanwhile, US stocks declined, with the Dow falling 1.6% and the S&P 500 losing 1.0%, amid concerns about AI spending.
  • A US security official suggested AI caused widespread outages, despite contrary evidence. Homeland Security Secretary Markwayne Mullin floated superintelligence as a possible explanation for disruptions affecting airlines, banks, and Microsoft. The organizations involved cited conventional technical problems, and no evidence has established that autonomous AI caused the incidents.
  • Amazon's CEO warned against slowing America's data center expansion. Andy Jassy argued that reducing AI infrastructure investment could allow China to establish global AI standards, even as US communities increasingly oppose large data center projects.
  • Anthropic reportedly raised its IPO expectations. The AI company is targeting a higher share price ahead of its expected public market debut later this week, amid weakness in tech stocks.
Advertisement

🍪 TOP TREATS TO TRY

  • GPT-6.2 Astra and Sol from OpenAI: OpenAI's newest models handle complex professional projects and everyday AI workflows. Both are available in ChatGPT Work, Codex, and the API for eligible users, with Astra priced at $8 per million input tokens and $40 per million output tokens.
  • Google Tasks is now available in Gemini: Google's former Spark assistant is now Tasks, with free access rolling out in selected countries. Users can delegate recurring research or schedule-related work, while additional US features include business calls and Google Pay checkout.
  • ChatGPT for Sales from OpenAI: OpenAI introduced a dedicated Sales plugin and over 30 partner integrations. Sales teams can research prospects, prepare customer presentations, and update CRM records through connected business applications.
  • Microsoft MAI Code 2 and Code 2 Flash: Microsoft released coding models for complex development and routine fixes at lower costs. The models form part of Microsoft's larger push into specialized AI, which includes cybersecurity and image generation models.
  • Astra for Law from OpenAI: OpenAI updated its legal research configuration using GPT-6.2 Astra. It passed every required check on 53.8% of questions in a recent legal research evaluation, up from about 46% in September. ChatGPT Enterprise customers can request access.
  • Suno Visuals creates videos for songs: Suno introduced AI-generated music videos that turn songs into 15-second clips, which can be combined into longer videos. Video generation requires additional paid credits.
  • RankSpot brings pay-as-you-go SEO to agents: A relaunched SEO tool that connects with ChatGPT or Claude through MCP, a standard for connecting AI agents to external services. It can research search results, generate articles, publish content, and support backlink outreach.
  • Tone AI brings Jordan Belfort's sales techniques to AI voices: The startup offers AI sales voices trained on his persuasion techniques. The company says it has raised around $10 million and plans to release an API.
  • Reflection AI Laser introduces restricted cybersecurity access: Reflection AI introduced a cybersecurity tool through a restricted-access program for vetted organizations. The company is limiting access to advanced security capabilities because of concerns about their potential use in cyberattacks.
Advertisement

🏢 Big Tech & Major Companies

💼 AI Productivity, Labor & Economics

  • Alphabet acknowledged that AI companies failed to convince some voters about data centers. Alphabet executive Ruth Porat said tech companies struggled to communicate the benefits of their massive infrastructure investments amid concerns about affordability and local construction impacts.

🏛️ AI Policy, Governance & Safety

Advertisement

Previous Around the Horn Digests

Catch up on everything you missed:

Advertisement

That's a Wrap

That's 21 AI stories and product releases in today's roundup. If you made it this far, congratulations. You now have more context on the AI industry than some of the people testifying about it.

For a shorter daily update, subscribe to The Neuron. We read the news and send you the stories worth your attention.

See you tomorrow.

P.S. Know someone who follows AI as closely as you do? Forward this digest and tell them to subscribe here.

Marianne Sison

Marianne is a technology analyst with nearly five years of experience reviewing collaborative work management solutions. She helps businesses identify the right tools and apply best practices to streamline workflows and improve project performance. Her insights on project management and unified communications appear in publications like Project-management.com, TechRepublic, and Fit Small Business.

The Neuron Logo

Don't fall behind on AI. Get the AI trends & tools you need to know. Join 700,000+ professionals from top companies like Microsoft, Apple, Salesforce and more.

Property of TechnologyAdvice. © 2026 TechnologyAdvice. All Rights Reserved

Advertiser Disclosure: Some of the products that appear on this site are from companies from which TechnologyAdvice receives compensation. This compensation may impact how and where products appear on this site including, for example, the order in which they appear. TechnologyAdvice does not include all companies or all types of products available in the marketplace.

Stay in the loop

Get notified when we publish new articles.