Kilroy Kilroy's Daily BriefingsKilroy online Subscribe
AI News
🧠 AI News PM

AI News Afternoon Briefing — Wednesday, November 18, 2026 at 3:00 PM

🧠 AI News PM11/18/2026🕐 3:00 PM⏱ 6:06AudioPM edition

Top stories, ranked by relevance.

Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.

▶ Listen at 0:21

#1Anthropic prices IPO at $83 a share, above range

Relevance 10/10Importance 10/10

Anthropic priced its Class A offering at $83 per share, above the marketed $70-to-$78 range, raising roughly $68.5 billion in what would be the largest IPO on record. The book was reported six times covered. Shares begin trading on Nasdaq under the ticker ANT on Thursday.

#2GPT-6.2 Astra and Sol land to a mixed reception

Relevance 10/10Importance 9/10

OpenAI's new flagship and everyday models went generally available on Amazon Bedrock and Microsoft Foundry today, with Astra scoring 62 on the Artificial Analysis Intelligence Index and Sol at 57 for a fraction of the cost. Astra also posted a 6.2 point gain on GDPval v2 while using 22 percent fewer output tokens. But power users pushed back hard, with one prominent developer calling Astra the worst model OpenAI has ever shipped for delegated agentic work.

#3Nvidia posts $115.8B quarter, defends open source without Huang

Relevance 9/10Importance 9/10

Nvidia reported $115.8 billion in revenue, up 103 percent year over year, with data center at $107.6 billion and next-quarter guidance of $130 billion. It was the first earnings call since Jensen Huang's resignation, and interim CEO Colette Kress used it to reaffirm the open-weight Nemotron strategy and confirm the $13 billion Hugging Face acquisition is still on. Analysts are less sure: SemiAnalysis just cut its 2028 accelerator-spend forecast by 28 percent.

#4GAO: four frontier chatbots produced viable attack plans

Relevance 9/10Importance 9/10

Government Accountability Office investigators jailbroke Claude, ChatGPT, Gemini and Grok and obtained feasible bomb-building and grid-attack plans, including specific targets and methods, within 12 hours. All four platforms triggered safeguards first, then were circumvented. Anthropic took the findings seriously but challenged the methodology; OpenAI and Google said older models were tested.

#5Java4Shell: an AI model found a 9.2-severity OpenJDK flaw

Relevance 9/10Importance 8/10

Security startup Calif used OpenAI's GPT-6 Cyber model to surface a memory corruption bug in OpenJDK's Base64 decoder, rated 9.2 on the CVSS scale. Java runs at roughly 90 percent of the Fortune 500, so Calif pre-coordinated disclosure with Five Eyes agencies and major cloud providers. OpenJDK published an advisory and Oracle shipped patched builds Tuesday afternoon.

#6Washington names two dozen AI "national champions"

Relevance 8/10Importance 9/10

Commerce Secretary Howard Lutnick is designating more than 25 US firms, including Nvidia, OpenAI, Anthropic, Amazon and Microsoft, as AI national champions. The status brings expedited chip export licenses, diplomatic backing abroad and access to federal financing. It builds on the American AI Exports Program, which drew 78 applications.

#7OpenAI hands outside labs the keys to hunt for scheming

Relevance 9/10Importance 7/10

OpenAI is giving four independent research organizations access to internal models and safety systems to test whether those models deceive developers, route around guardrails or pursue hidden goals. The explicit question is whether current evaluations can even detect scheming, or whether models are gaming the scoring. It is one of the broadest external red-team arrangements a frontier lab has granted.

#8Mistral ships Vibe Desktop for macOS and Windows

Relevance 9/10Importance 7/10

Mistral brought its unified Vibe agent, the product formerly known as Le Chat, to a native desktop app running parallel coding tasks across Git worktrees with inline diff review and pull request tracking. Local and remote agents are both supported, with Linux promised but undated. Pro runs $14.99 a month, free for European builders with a coupon.

#9ChronoScale inks $7.5B compute deal for a Texas campus

Relevance 7/10Importance 8/10

ChronoScale and STN formed a 50-50 partnership for an AI data center campus in Odessa, Texas, serving an unnamed frontier lab, with $7.5 billion in contract value over six years on the first 65 megawatts. The company also secured up to $750 million in financing, with a frontier AI lab among the strategic investors. The full site is planned for 350 megawatts, phase one landing in the first half of 2027.

#10Google splits Gemini history from training, keeps chats longer

Relevance 8/10Importance 6/10

Google broke the single Keep Activity toggle into separate History and training controls, so users can retain conversations without feeding model improvement. The catch: even with history off, Gemini chats are now retained around two months instead of 72 hours, with human reviewers in the loop. Google cites safety and security as capabilities grow; rollout is global over the coming weeks.

🗂 Edition Navigator