What happened

OpenAI: GPT-6 Cyber, kernel exploits and more

On 5 November OpenAI announced GPT-6 Cyber, a model it says is trained to find and validate exploitable vulnerabilities. It reports 48.5% on its ExploitGym test against 32.1% for GPT-5.6 Cyber and says it helped find 81 high-severity Firefox flaws; OpenAI rates it “Critical” for cybersecurity under its Preparedness Framework, and access starts with approved Daybreak Red customers. (full article) A day earlier OpenAI described an agent workflow that reproduced 1,724 Linux kernel vulnerabilities and produced 948 local privilege-escalation exploits for about $1.2 million in API credits, and says it sent hundreds of reports and patches to maintainers. (full article) OpenAI also plans a new security sprint with the firm Cathedral and says it will more than quadruple critical bug bounty rewards. (full article)

Images 2.6 launched in ChatGPT, Codex and the API with animated GIFs, and OpenAI says it generates images 50% faster. (full article) GPT-Rosalind-6, a biology model that shares weights with GPT-6 Astra, is offered only to approved scientists and partners. (full article) Four agreements were signed between Abu Dhabi entities and OpenAI, with no amounts given (full article), and OpenAI named former Army Secretary Dan Driscoll as Senior Counselor for National Security Modernization. (full article)

Anthropic: S-1/A, Ultra plan, watermark

On 2 November Anthropic filed an amendment to its S-1 that adds third-quarter and nine-month results and its total compute commitments; the price range and share count are still blank. (full article) Its pricing page lists Claude Max at $100 and Ultra at $200 a month, with no 5-hour limit on Ultra for verified accounts. (full article) Anthropic says text from Claude Mythos 5, Opus 4.7 and Sonnet 4.6 now carries a text watermark, although its own support table still marks them as not watermarked. (full article) It also published FlashProve, a research prototype that proves which model produced an output (full article), and put message threads in beta in the Messages API. (full article)

Apple: China and Siri AI

Apple said on 31 October that Apple Intelligence reaches users in China in December with iOS 27.3, with Alibaba’s Qwen available in Siri, Writing Tools and Visual Intelligence; Siri AI is not included. (full article) On 3 November Apple opened a Siri AI developer preview in the 27.3 betas: any developer can expose an app’s content and actions through App Intents and teach Siri multi-step skills. Apple says it ships to customers in a few months and is unavailable in the EU for now. (full article)

Google: Omni 1.2 Flash, classrooms, chemistry

Gemini Omni 1.2 Flash makes native clips of up to 15 seconds and lists $0.07 per second for 720p video, 30% below Omni 1.1. (full article) Google said Gemini and Search will be off by default for K-12 student accounts until administrators turn them on (full article), and DeepMind added Cheminformatics Skills to Antigravity with free top-tier use for 10,000 scientists for six months. (full article)

Models and open weights

  • Inkling 2 Small (Thinking Machines Lab): 594B total and 17B active parameters with image input, available now with closed weights; the lab says Apache 2.0 weights follow next week after safety testing. (full article)
  • Kimi K3.1 Pro (Moonshot): the weights are now on Hugging Face under a custom license we have not read. (full article)
  • ERNIE 5.5 Flash Base (Baidu): a 280B base checkpoint with 14B active parameters, under Baidu’s own license. (full article)
  • Hydrogen-1.0 (Deep Principle): a 753B science model post-trained from GLM-5.3, with MIT-licensed weights. (full article)
  • SenseNova-Loop-9B-Preview: a looped model that reuses 24 layers twice, under Apache 2.0. (full article)
  • Krea 3: the technical report says an edit takes about 8 seconds, by Krea’s own figures. (full article)

Security and research

CrowdStrike added an AI Kill Switch to Falcon Guardian that tells a connected AI gateway to block an agent’s access to its model. (full article) A DeepSeek team open-sourced dsh-superagent, a prototype that is not an official DeepSeek product. (full article) Cohere added third-party open models, including DeepSeek V4.1 Flash and GLM 5.3 Flash, to Model Vault. (full article) Four MIT researchers report that GPT-6 Astra edited the weights of small networks without gradient descent, raising Qwen3.5-0.8B on MATH-500 from 7.9% to 48.1% but not improving Gemma; they call the results early. (full article)

Business, chips and policy

  • Marvell and Lumentum: an all-stock deal with an implied enterprise value of $115 billion, expected to close in the first half of 2028. (full article)
  • AMD: third-quarter revenue of $13.4 billion, Data Center up 97% to $8.5 billion, fourth-quarter guidance of about $17 billion. (full article)
  • Cohere: initial close of a Series E targeting up to CAD 7 billion, and a commitment of more than C$4 billion in Canada by 2031. (full article) (full article)
  • fal: Series E and F rounds with no amounts disclosed; press reports put the Series F at $1.3 billion. (full article)
  • Policy: President Trump posted that he is “100% in favor of Super Intelligence” and will work with states; the post announces no policy, order or rule. (full article)

Key details

  • GPT-6 Cyber: 48.5% on ExploitGym against 32.1%; approved Daybreak Red customers first; all figures are OpenAI’s own.
  • Anthropic S-1/A: third-quarter and nine-month results added; price range and share count still blank.
  • Siri AI: developer preview in the iOS 27.3 betas; customer shipping in a few months; unavailable in the EU for now.
  • Inkling 2 Small: closed weights now; Apache 2.0 weights are a plan for next week, not yet a fact.
  • Kimi K3.1 Pro: weights are on Hugging Face under a custom license.
  • Benchmarks: nearly all performance figures this week come from the companies that released the products.
  • Next week: Inkling 2 Small weights, Anthropic’s price range, Siri AI timing and Apple Intelligence in China in December.

Why it matters

Two threads ran through the week. Security capability is being released in controlled steps: GPT-6 Cyber and GPT-Rosalind-6 go only to approved customers, and OpenAI’s kernel run shows how much work an agent can do for about $1.2 million, which raises the pressure on maintainers to patch quickly. And open weights keep arriving on promises: Kimi’s weights shipped after being promised, while Inkling 2 Small’s are still a plan.

For builders, the practical items are Siri AI’s App Intents preview, Omni 1.2 Flash’s per-second price, and the licenses attached to each new open model.

Disclosure: Claude, made by Anthropic, was one of the AI tools used to research and draft this article, and Anthropic is the subject of several items in it; Anthropic competes with OpenAI and Google. The voice in our video version is made by ElevenLabs. Most figures are the companies’ own. Nothing here is investment advice.

Sources

  1. GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents Primary source , OpenAI
  2. A thousand fewer paths to root Primary source , OpenAI
  3. Preparing our next security sprint to harden OpenAI Primary source , OpenAI
  4. Introducing Images 2.6 Primary source , OpenAI
  5. GPT-Rosalind-6 System Card Addendum Primary source , OpenAI Deployment Safety Hub
  6. Khaled bin Mohamed bin Zayed meets Sam Altman and witnesses signing of agreements with OpenAI (Abu Dhabi Media Office, 2 November 2026) Primary source , Abu Dhabi Media Office
  7. OpenAI appoints Dan Driscoll as Senior Counselor for National Security Modernization Primary source , OpenAI
  8. Anthropic, PBC — Form S-1 registration statement (SEC EDGAR) Primary source , U.S. Securities and Exchange Commission
  9. Claude pricing (Anthropic) Primary source , Anthropic
  10. Claude Platform release notes (3 November 2026 entry) Primary source , Anthropic
  11. Progress on provable inference for large language models Primary source , Anthropic
  12. Apple Intelligence is coming to China (Apple China newsroom, press release dated 31 October 2026, in Chinese) Primary source , Apple
  13. Siri AI is now open to all developers (Apple Developer News, 3 November 2026) Primary source , Apple
  14. Introducing Gemini Omni 1.2 Flash Primary source , Google
  15. Designing responsible AI for learning Primary source , Google
  16. Agentic cheminformatics with Google Antigravity Primary source , Google DeepMind
  17. Introducing Inkling 2 Small Primary source , Thinking Machines Lab
  18. moonshotai/Kimi-K3.1-Pro (Hugging Face model repository and API listing) Primary source , Hugging Face / Moonshot AI
  19. Introducing ERNIE 5.5 Flash Base Primary source , Baidu ERNIE Team
  20. Introducing Hydrogen-1.0 (Deep Principle blog) Primary source , Deep Principle
  21. SenseNova-Loop-9B-Preview (Hugging Face model card) Primary source , SenseNova
  22. Krea 3 Technical Report (Krea blog, 3 November 2026) Primary source , Krea
  23. Introducing AI Kill Switch: Taking Control When AI Agents Go Rogue Primary source , CrowdStrike
  24. deepseek-ai/dsh-superagent-prototype (GitHub) Primary source , DeepSeek
  25. The world's leading open models, fully under your control (Cohere blog, 3 November 2026) Primary source , Cohere
  26. Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights (Takehi, Teoh, Zhang, Khattab, 2 November 2026) Primary source , MIT CSAIL authors
  27. Marvell and Lumentum to Combine, Creating a Category Leader in AI Connectivity (Marvell press release, 2 November 2026) Primary source , Marvell
  28. AMD Reports Third Quarter 2026 Financial Results Primary source , AMD (GlobeNewswire)
  29. Who Gets to Build the Frontier for AI Primary source , Cohere
  30. Cohere advances C$4 billion commitment to strengthen Canada's AI leadership and significantly expand its national workforce (Cohere blog, 30 October 2026) Primary source , Cohere
  31. fal blog: Series E and Series F announcement Primary source , fal
  32. Donald J. Trump post on Truth Social, 6 Nov 2026, 7:01 AM ET (not opened directly by Deyron Labs) Primary source , Truth Social