AI news for Friday, 23 October 2026

The day’s AI news in one list, the biggest stories first. Each story links to its short read and to the outlets that reported it.

1. 1 · Policy & Drama
   OpenAI trains advanced AI models offline after hacking incidents

   OpenAI is now training its most advanced AI models in an offline environment using cached and indexed web content instead of live internet access.

   Reported by The Verge and 1 more

2. 2 · Models & Releases
   Block releases Dessa, an AI model trained on human speech

   Block announced Dessa, an AI speech model trained entirely on human conversation that processes audio directly rather than converting speech to text first.

   Reported by The Verge and 2 more

3. 3 · Hardware & Compute
   Google Cloud Launches Agent Compute Engine for AI Agents

   Google Cloud announced Agent Compute Engine, a compute service that runs each AI agent in its own hardware-isolated microVM with zero stored credentials.

   Reported by MarkTechPost and 1 more

4. 4 · Funding & Deals
   Pathos AI raises $237M for cancer drug development

   Pathos AI, a startup developing cancer drugs using an artificial intelligence platform called Foundry, raised $237 million in funding at a $3 billion valuation, with pharmaceutical giant Bristol Myers Squibb leading the investment round.

   Reported by SiliconANGLE

5. 5 · Hardware & Compute
   Amazon's Trainium redesigned for easier AI training

   Amazon's accelerator chip for AI training has been redesigned to work more like the industry-standard approach used by competitors, eliminating the need for researchers to rewrite their code when switching from other platforms.

   Reported by SemiAnalysis

6. 6 · Robotics
   Generative Bionics to launch Physical AI Data Factory in Italy

   A company is building a large physical facility in Europe to generate training data for humanoid robots by having them perform real industrial tasks in simulated environments.

   Reported by The Robot Report

7. 7 · Policy & Drama
   OpenAI strengthens security for frontier model training

   OpenAI has implemented new security measures to prevent frontier models under development from escaping their training environments and gaining unauthorized access to the live internet or other systems.

   Reported by OpenAI

8. 8 · Agents & Products
   Google Research develops AI tool to detect tuberculosis from X-rays

   Google Research has developed AI tools that can detect tuberculosis from chest X-rays with accuracy matching expert radiologists, addressing a critical gap in diagnostic access for remote and resource-limited areas.

   Reported by Google Research

9. 9 · Policy & Drama
   Intel receives waiver for foreign-made robots in US fabs

   The US government implemented a ban on certain foreign-made robots through the Federal Communications Commission, which created confusion about which devices were actually restricted.

   Reported by Ars Technica

10. 10 · Policy & Drama
    Now Google says it might use your Gmail and Google Photos data to target ads

    Google has updated its privacy policy to allow targeted ads based on personal data from Gmail, Google Photos, and Drive when users enable the company's personalized AI features.

    Reported by The Verge

11. 11 · Robotics
    inDrive plans robotaxi tests in Kazakhstan in 2027

    Ride-hailing company inDrive has signed an agreement with autonomous driving firm Pony.ai and Nvidia to test robotaxis in Kazakhstan starting in the first half of 2027, with initial rides operated by a safety operator before moving toward…

    Reported by TechCrunch

12. 12 · Agents & Products
    Zoom shifts focus to AI data platform

    Zoom is repositioning itself from a video conferencing company to an artificial intelligence data platform that captures and analyzes conversational data from meetings, calls, chat, and other interactions.

    Reported by SiliconANGLE

13. 13 · Robotics
    Hottest toy for Christmas? Nvidia CEO plays with adorable Microduck robots, 20,000 sell

    A video of a company CEO playing with small robotic devices called Microducks during a conference keynote generated significant online attention.

    Reported by TechCrunch

14. 14 · Agents & Products
    Enterprise AI shifts toward cost control and hybrid infrastructure

    Enterprise artificial intelligence is entering a phase where cost considerations are reshaping business decisions about where AI models run and which models to use.

    Reported by SiliconANGLE

15. 15 · Agents & Products
    OpenAI scales Habitat storage to serve over 1 billion ChatGPT users

    OpenAI describes how it scaled its online storage platform, called Habitat, to serve over 1 billion users across 40 geographic regions by optimizing its use of Azure Cosmos DB.

    Reported by OpenAI

Every story, by topic