AI Daily · Nov 19, 2026 · ThuAI Daily
ICREAT AI NEWS
Google DeepMind releases Gemini 4 Neon model
Google DeepMind releases Gemini 4 Neon, focused on end-to-end task completion for complex software engineering, knowledge work, and scientific workflows, ranking 3rd on the Artificial Analysis Intelligence Index with a score of 61.
Model releases/updates
OpenAI releases GPT-6.2 Astra and Sol; Astra's intelligence index of 62 trails Fable 5.2
OpenAI released GPT-6.2 Astra and Sol; Astra scores 62 on the Artificial Analysis Intelligence Index, trailing Fable 5.2's 65, while Sol matches September's Astra at about 15% of the cost per task.
Microsoft unveils seven in-house MAI models at Ignite, replacing external models at lower cost
Microsoft unveiled seven in-house MAI models at Ignite in one go, covering code, security, image, speech, routing and retrieval, with a focus on cost-effectiveness. Microsoft says MAI Code 2 averages 98% of GPT-6.1 Sol's performance on four benchmarks at 40% of the cost, and 96% of Claude Opus 5.5's performance at 5% of the cost.
Product releases/updates
Google DeepMind opens Genie 4 world model to AI Ultra subscribers
Google DeepMind opened a new Project Genie experience to eligible Google AI Ultra subscribers, offering the Genie 4 world model for the first time.
Mastra releases new capabilities for open-source software factory Mastra Factory
Mastra released new capabilities for its open-source software factory Mastra Factory, supporting integrations with GitHub, GitLab, Linear, Jira and other sources, customizable workflow stages such as Triage, Plan, Review and Done, and built-in cloud coding agents and E2B-based microVM sandboxes.
Industry news
Wiz Research discloses two unauthenticated remote code execution vulnerabilities in NGINX HTTP/2
Wiz Research, together with Ikotas Labs, Metnir, Rapid7 and Xint, discovered two unauthenticated remote code execution vulnerabilities in NGINX HTTP/2, CVE-2026-123633 and CVE-2026-123007, both rated Critical with a CVSS 4.0 score of 9.2.
FBI warns AI hacking tool ARTEX can breach networks in under three hours
In a FLASH alert dated November 13, the FBI said hackers are using the open-source AI attack framework ARTEX to target at least ten industries, with some intrusions going from first probe to privileged access in under three hours.
GAO tests bypassing ChatGPT, Claude, Gemini and Grok guardrails, obtaining attack plans
At the request of Senator Maggie Hassan, the U.S. Government Accountability Office (GAO) tested ChatGPT, Claude, Gemini and Grok in May, using methods not disclosed in the report to bypass safety guardrails and obtaining, within 12 hours, two detailed plans for mass-casualty attacks and two plans to attack the power grid and cause major economic damage.
Nvidia stops shipping H200 AI chips to China
Nvidia interim CEO Colette Kress said on the earnings call that the company no longer ships licensed Hopper data center products to China, and that its guidance does not include such products. She did not explain the reason for halting shipments or whether the company will continue doing business in China in the future. Reuters previously reported that about 600,000 H200 chips were stranded in Hong Kong because Beijing had not approved large-scale imports.
How Cloudflare WAF defends against Java4Shell (CVE-2026-114938)
Cloudflare added rules to its managed ruleset to detect malformed Base64 input related to Java4Shell (CVE-2026-114938), available to Pro, Business and Enterprise plans and enabled by default with a Log action.
Reuters exposes Meta internal chats: Muse agent alleged to read private messages, leak data across accounts
Internal Meta chats exposed by Reuters show employees warning during testing of the Muse agent that it could read encrypted private messages users did not expect to be viewed, mix up conversations between different users, and even output unfamiliar bank account balances and Bitcoin balances.
SEC opens civil investigation into Nvidia over China-related chip sales disclosures
According to The Wall Street Journal, the SEC has opened a civil investigation into Nvidia to determine whether it violated disclosure rules and misled investors, focusing on Nvidia's chip sales in Asia. Nvidia changed its revenue disclosure last year and no longer separately lists sales to Singapore, while Megaspeed, a Singapore-headquartered data company, has been under investigation by the U.S. Commerce Department since late 2024 over allegations that it used Nvidia chips to build AI data centers in Southeast Asia and that some chips flowed into China.
Briefs
- Anthropic reportedly prices IPO at $83 per share, above prior rangeTechCrunch · AIFollow-up
- Study shows Google AI summaries reduce clicks to external websites by about 40%Hacker News AI
- Axiom Math uses AI to automatically formalize the classification of finite simple groups theorem and the Stacks projectHacker News AI
- Open WebUI launches Open Terminal, giving every model an operable Linux environmentHacker News AI
- Fed reviews AI infrastructure financing for the first time, finds over $3 trillion in off-balance-sheet commitmentsHacker News AI
- Bank CEOs and NSA tell Congress AI-driven patch fixes caused this week's system outagesHacker News AI
- OpenAI says its models completed about a year's worth of chip optimization work on top of AMD's existing engineeringThe Decoder
- RadixArk releases DaisyRose-27B, post-trained with reinforcement learning on Qwen3.8-27BHugging Face Blog
- Reson8 open-sources 0.5B speech recognition model Rumble-2-Flash, supporting 40+ languagesHugging Face Blog
- Databricks launches Unity Gateway Insights, using Genie to analyze agent traces and pinpoint token wasteDatabricks Blog