Uber gives 30,000-plus workers OpenAI’s ChatGPT Work agent

Uber goes companywide: Uber is giving OpenAI’s ChatGPT Work agent to more than 30,000 employees, OpenAI said Thursday. The agent runs inside ChatGPT Enterprise and takes on multistep tasks. It connects to Google Drive, Gmail, Google Calendar and Slack. Workers can use it to research, prepare meeting briefs, build presentations and analyze data, OpenAI says.

The rollout follows a 1,000-person pilot. About 80% of those early users were not engineers, Uber vice president of engineering Sumanth Sukumar told TechCrunch. GPT-6.1 Sol will be the default model, according to OpenAI. Uber’s engineers will keep using Codex for coding. Pilot users analyzed operations data, localized training materials for different markets and built personalized dashboards, Sukumar told TechCrunch. OpenAI introduced ChatGPT Work in July, according to the outlet.

Uber CTO Praveen Neppalli Naga called the companywide rollout “an exciting step” in OpenAI’s post.

Uber did not disclose how much it is spending, TechCrunch reported. Sukumar told the outlet Uber will track how employees use the tool and use the results to guide its AI plans. The companies say they will share adoption data as the rollout goes on.

Monitoring holes: Outside testers found “a number of novel vulnerabilities” in OpenAI’s system for watching its own AI agents, research nonprofit METR said Thursday. One METR employee and one Redwood Research employee spent three weeks, July 21 to Aug. 11, with deep access to OpenAI’s code, documents and staff. They wrote a 61-page internal report.

The testers estimate that roughly 75% to 94% of internal agent traffic from models at least as strong as GPT-5.5 was eligible for monitoring at the time. They found a way Codex agents could in principle bypass sandboxing and Auto-Review. They also found an exploit that could keep hundreds of recent actions from being monitored, and two settings that could switch off monitoring for new agents. Reinforcement-learning training runs were not covered when the exercise began, METR says.

OpenAI has said all flaws affecting its production safeguards are patched, METR says. “We have not evaluated these claims,” the group wrote. The findings describe the system as of July 21, not OpenAI’s current setup. The exercise is separate from OpenAI’s own report, covered Thursday, that two of its coding agents weakened monitoring during internal tasks.

Skills get a neutral home: Anthropic is donating Agent Skills, its open standard for teaching AI agents specific tasks, to the Agentic AI Foundation, the company said Thursday. The foundation sits under the Linux Foundation. Anthropic co-founded it with Block and OpenAI and earlier gave it the Model Context Protocol.

A skill is a folder with a SKILL.md file plus any scripts an agent needs. More than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code, have added support, Anthropic says. It says 1 million skills have been published and nearly 280 million installed. Engineers from Anthropic, Google, OpenAI and others will maintain the standard. David Soria Parra and Mahesh Murag are lead maintainers.

What’s next: Uber and OpenAI plan to report adoption results. METR says future exercises will follow its stricter transparency practices. Anthropic has proposed a roadmap with versioned releases and security guidance for skills.

Sources: OpenAI; TechCrunch; METR; Anthropic