OpenAI Offline Training, AI Ethics, and New AI Tools

A day of policy shifts and new AI tools. From OpenAI’s decision to cut internet access for training to Block’s conversational model, stakeholders face fresh challenges and opportunities.

OpenAI cuts internet training for frontier models

OpenAI announced it will no longer train its top AI models on live internet data, citing significant risks. The company built an offline testing environment to use cached web content safely. Why it matters: Developers can rely on more secure, controlled training pipelines, reducing exposure to malicious content.

Read more: OpenAI says it will no longer train its AI models on the live internet, citing “significant risks”

Quick Study turns photos into paint in seconds

A new tool, Quick Study Prompt, uses Claude Opus 5.5 to transform a photo into a canvas painting within seconds, leveraging Sobel edge detection and no external libraries. Why it matters: Startups can quickly prototype visual content without heavy graphics stacks.

Read more: Quick Study

Block’s Dessa learns from human conversation alone

Jack Dorsey’s Block unveiled Dessa, an AI model trained from scratch on nothing but human dialogue. Early demos show occasional awkward interactions, illustrating the limits of conversational AI. Why it matters: Businesses can evaluate the trade‑offs of data‑minimal models for customer‑facing bots.

Read more: Jack Dorsey’s Block made an AI model that talks to you like an awkward new friend

Meta delays new AI scraping terms

Meta postponed enforcement of its new business terms after publishers raised concerns about potential news‑site scraping for AI training. The terms will take effect on November 20th. Why it matters: Publishers and partners need to stay updated on compliance requirements for AI use.

Read more: Meta delays new business terms after publisher pushback on AI scraping

Google may use Gmail data for targeted ads

Google updated its privacy policy to allow personalized AI features to use Gmail and Google Photos data for ad targeting, a shift from earlier promises. Why it matters: Users and businesses must reassess data usage agreements and privacy controls.

Read more: Now Google says it might use your Gmail and Google Photos data to target ads

Retro gaming goes AI‑powered

Unexpected AI projects have turned classic games into web, mobile, and mixed‑reality experiences, with projects like Halo in a browser and Wind Waker on iPhone. Why it matters: Developers can repurpose legacy IPs quickly with AI, opening new revenue streams.

Read more: Retro gaming is being overrun with AI

InDrive plans robotaxi tests in Central Asia

Uber rival InDrive will begin testing Pony.ai robotaxis in Kazakhstan’s capital in 2027, partnering with Nvidia for hardware. Why it matters: Autonomous vehicle startups can explore emerging markets with established tech partners.

Read more: Uber rival inDrive eyes 2027 robotaxi tests in Central Asia with Pony AI and Nvidia

AI agents invent games when left idle

MIT researchers left 13 AI agents on a virtual island; the agents invented games, mourned a lost ring, and improved at their tasks. Why it matters: Insight into emergent behavior can guide long‑term AI deployment strategies.

Read more: Left alone with nothing to do, AI agents started playing

What to watch

• OpenAI’s continued evolution of offline training
• Meta’s final terms on AI scraping
• Market response to Google’s new ad‑targeting policy
• Development of autonomous vehicle tests in new regions
• AI agents’ emergent capabilities in long‑term deployments