AI & Automation Briefing - October 24, 2026
Uber Rolls Out ChatGPT Work to Its Entire 30,000-Person Workforce
Uber has expanded its OpenAI partnership to give all 30,000 global employees access to ChatGPT Work, the enterprise-focused AI agent OpenAI launched in July. The company started with a 1,000-person pilot earlier this year, drawing participants from finance, operations, and product. Roughly 80% of those early users were non-engineers, who applied the tool to tasks like analyzing operational data, localizing training materials, and building custom dashboards. Uber is not prescribing specific use cases for the full rollout. Instead, it plans to monitor which workflows employees build and use those findings to guide broader AI deployment. Earlier this year, Uber also ran an experiment pairing 30 AI engineers with cross-functional teams for two weeks to automate parts of their jobs, a format the company called AI pods. Uber did not disclose the cost of the expanded rollout.
MIT Researchers Find AI Agents Invent Their Own Challenges When Left Without a Task
MIT researchers placed 13 AI agents, each running on Claude Opus 4.7, on a simulated desert island for 30 hours with no assigned objectives. The agents began setting self-directed challenges, a behavior the researchers describe as play. When subsequently tested on specific tasks, such as moving as many objects as possible to the island's highest point, the agents that had explored freely outperformed identical agents with no prior island experience. Six of the 13 agents independently discovered they could throw objects farther by releasing them mid-jump, a technique the researchers had not intentionally built into the environment. The team argues this unstructured exploration offers an alternative training path for AI agents, particularly for future robotics applications, where skills accumulated through open-ended discovery may complement conventional goal-defined training methods. The study is currently available on arXiv as a preprint pending peer review.
Research Shows LLM Agents Leak Sensitive Data Even When Individual Queries Appear Safe
A new study called Agent Mosaic identifies a confidentiality risk in agentic LLM systems where combining multiple individually harmless data points exposes sensitive conclusions. Researchers tested this across simulated and live agents and found that leaks occurred even under routine queries and persisted even when explicit privacy safeguards were in place. For operations leaders deploying AI agents across workflows with access to client, team, or business data, this highlights a gap that access controls alone may not close.