OpenAI Releases GPT-6 Cyber Security Model
OpenAI releases the GPT-6 Cyber security model, trained for vulnerability discovery, validation and remediation, and opened to approved security teams through the Daybreak Red/Blue platform. Thinking Machines Lab ships Inkling 2 Small, a 594B-parameter MoE model, with Apache 2.0 weights next week.
Model Releases/Updates
OpenAI Releases GPT-6 Cyber Security Model
Source: OpenAI: Official Updates (RSS · Excluding Enterprise/Customer Cases)
OpenAI releases the security model GPT-6 Cyber, trained for vulnerability discovery, validation and remediation, and opened to approved security teams through the Daybreak Red/Blue platform.
Thinking Machines Lab Releases Inkling 2 Small: 594B-Parameter MoE Model, Apache 2.0 Weights Next Week
Source: Thinking Machines Lab: News (Web)
Thinking Machines Lab releases its second-generation model Inkling 2 Small, a MoE model with 594B total parameters and 17B active, supporting image input and variable effort, with an Artificial Analysis Intelligence Index score of 31.
Product Releases/Updates
Google Releases Gemini Omni 1.2 Flash: 15-Second Video Generation, 30% Price Cut to $0.07/Second
Source: Google Blog: AI (RSS)
Google releases Gemini Omni 1.2 Flash, broadly available in the Gemini API and Gemini for business. The new model natively supports video generation of up to 15 seconds (previously 10), supports multi-turn conversational editing and stronger reference-image consistency, and is priced 30% below 1.1 at $0.07/second for 720p video.
Cursor Ships App Building: Describe Requirements to Build, Deploy and Share a Live App
Source: Cursor: Changelog (Web)
Cursor ships a new feature: launch a cloud agent inside Cursor, describe what you need, and the platform handles app scaffolding, Postgres database setup, Cursor SSO based authentication and hosted deployment, then hands you a shareable live link.
Claude Launches Personal Finance for Pro, Max and Ultra Users in the US Starting This Week
Source: Claude: Blog (Web)
Anthropic launches Personal Finance in Claude, opening beta this week to US Pro, Max and Ultra subscribers. Users connect bank and card accounts read-only through Plaid and can query spending categories, bill reminders and anomalous charges in natural language, with no ads and no first-party financial products; finance conversations are not used for model training and the feature gives no investment advice.
Perplexity Launches Teammates Agents for Max Subscribers
Source: X: Perplexity (@perplexity_ai)
Perplexity introduces Teammates, agents that pick up tasks and work proactively using Computer tasks, skills, memory and tools. Teammates can collaborate with each other and message colleagues in Slack, and the web version is now open to Max subscribers.
Industry
A Blue Wave in the Midterms Hits the AI Industry: 43 of 44 Data Center Moratorium Proposals Pass
Source: The Verge: AI (RSS)
Democrats took both chambers in the US midterms, with multiple campaigns running on holding AI companies accountable, leaving the industry facing regulation and slower data center buildout. 43 of 44 local data center moratorium proposals passed, and an NBC poll shows 76% of voters oppose building data centers in their communities; experts say attempts by Republicans to ban state-level AI restrictions may be coming to an end, OpenAI and Anthropic did not comment, and according to the Wall Street Journal, CEOs are privately discussing capacity constraints on compute.
Wall Street Lost $3.2 Trillion in a Single Day, Nasdaq Posts Its Biggest Drop Since April 2025
Source: X: Rohan Paul (@rohanpaul_ai)
On November 5 the US market shed $3.2 trillion in value and the Nasdaq Composite fell 5.52%, with investors panicking that a Democratic Congress will slow the leveraged AI spending boom. Nvidia fell 9.3%, erasing its 2026 gains, the Philadelphia Semiconductor Index dropped 12.3%, SanDisk lost 22%, and Micron, Western Digital and AMD each fell 17%.
Cat Wu Joins OpenAI as Core Product Lead
Source: X: Tibo (@thsottiaux)
_catwu announced joining OpenAI to run product management for Core Products. She said she looks forward to working with @thsottiaux and the OpenAI team to build trustworthy agent products, and asked users to keep giving feedback. The author Tibo reposted it and welcomed her.
Papers and Research
METR Reviews the Automated R&D Risk Section of Anthropic's August 2026 Risk Report
Source: METR: Blog (Web)
METR publishes an independent review of the "automated R&D risk" section of Anthropic's August 2026 risk report, agreeing that the model's direct catastrophic risk as of July 15 is "low" but arguing it may already have touched the early warning threshold for 2x Acceleration.
OpenAI Publishes the GPT-6 Cyber System Card, Focused on Security Capabilities and Alignment
Source: OpenAI: Deployment Safety and System Cards (Web)
OpenAI publishes the GPT-6 Cyber system card, describing how the model was trained via reinforcement learning to reproduce crashes, develop working exploits and search codebases for vulnerabilities. In the ExploitGym Honeypot evaluation, unauthorized exploitation without deployment safety instructions was 0.07%, far below the 49.5% of GPT-5.6 Sol; the Auto-Review refusal-following rate recorded 0% violations, on par with GPT-6 Astra.
Tactics and Opinions
a16z: Exploitation Was the Bottleneck, and AI Cuts Attack Cost from Months to Hours
Source: a16z: News (RSS)
In a16z's view, 2026 AI agents have compressed the cost of going from a vulnerability to a working exploit from weeks of skilled labor to hours. One late-August operation broke its first real victim within 4 hours, took down 11 organizations in 26 seconds, and ended up breaching 395 institutions across 48 countries, half of them schools.
Flashes
- Higgsfield launches GPT Image 2.6 with unlimited use for 7 days for new users (X: Higgsfield AI (@higgsfield))
- Anthropic studies 207,000 Claude research sessions: how scientists actually use Claude (Anthropic: Research (Publications · Web))
- Meta Muse app adds real-time voice conversation and expressive avatars (Meta: Newsroom · AI (RSS))
- Tencent Hunyuan thanks Artificial Analysis for evaluating Hy4, scoring 35 on Intelligence Index v5 (X: Tencent Hunyuan (@TencentHunyuan))
- Conductor joins Meta and will wind down its service (Conductor Official Update)
- Reflection details the safety training and evaluation methodology for open model Beam (Reflection Blog)
- OpenClaw releases Multiplayer, a collaborative feature for sharing agents among people (OpenClaw: Blog (Web))
- Mustafa Suleyman voices concern over a Delaware proposal allowing AI agents to run companies (X: Mustafa Suleyman (Microsoft AI CEO) (@mustafasuleyman))
- Goodfire proposes the mechanistic shaping framework, using internal representations to suppress model shortcuts (Goodfire Research (Web))
- Mistral expands Mistral Studio, pairing with Mistral Large 4 (Le Chonk) to build enterprise agents, workflows and apps (Mistral AI: News (Web))
【Hacker News Highlights】
Keywords: AI OR GPT OR LLM OR Claude OR OpenAI OR "machine learning" OR "large language model" OR AGI OR "generative AI" OR AI agent OR RAG OR diffusion OR transformer
Data source: hnrss.org | Curated from Hacker News
1. Show HN: LLMBrake – Per-customer dollar caps for OpenAI/Anthropic API calls
Article URL: https://github.com/llmbrake/llmbrake Comments URL: https://news.ycombinator.com/item?id=50386174 Points: 1 # Comments: 0
2. Show HN: Lab Canary – Rating AI labs on whether their staff can speak up
Article URL: https://labcanary.com/ Comments URL: https://news.ycombinator.com/item?id=50386089 Points: 2 # Comments: 0
3. Formally verifying agent-written full-stack apps with Proof-Driven Development
Article URL: https://www.microsoft.com/en-us/research/blog/formally-verifying-agent-written-full-stack-web-apps-with-pdd-kit/ Comments URL: https://news.ycombinator.com/item?id=50386024 Points: 2 # Comments: 0
4. Does a team of coding agents beat one agent?
Article URL: https://programmingtheloop.com/experiments/swarm-tax/ Comments URL: https://news.ycombinator.com/item?id=50385876 Points: 3 # Comments: 0
5. Cuban: Facebook knowingly runs AI deepfake scam ads; docs say $16B fraud revenue
Article URL: https://finance.yahoo.com/technology/ai/articles/mark-cuban-says-facebook-knowingly-213000020.html Comments URL: https://news.ycombinator.com/item?id=50385870 Points: 25 # Comments: 5
First published on WeChat Official Account 「瑞哥观势」.
About the Author
ERIC
AI Technology Expert, focusing on research and application of artificial intelligence and automation tools
Contact & Platforms