OpenAI Launches Cloud-Based AI Agents for Team Collaboration in ChatGPT
OpenAI introduced "workspace agents" in ChatGPT — cloud-based AI assistants that work autonomously and can be shared across teams. The company also released a free Privacy Filter tool to protect sensitive data.
Cloud Agents Designed for Organizational Workflows
OpenAI has introduced "workspace agents"in ChatGPT — cloud-based AI assistants built for team collaboration. These agents are powered by Codex and can handle tasks typically performed by employees, from preparing reports and writing code to responding to messages.
Since the agents operate in the cloud, they can continue executing tasks even when the human user is offline. Teams create an assistant once and then share it across the organization through ChatGPT or Slack.
Why This Matters
AI tools have largely boosted individual productivity until now. OpenAI's workspace agents target a fundamentally different challenge: team-level processes that involve shared context, task handoffs between employees, and cross-team decision-making.
According to OpenAI's announcement, the agents can gather context from relevant systems, follow team processes, request approvals when needed, and coordinate work across different tools. This represents a shift toward automating entire organizational workflows rather than isolated tasks.
Sales Team Use Case
OpenAI demonstrated the feature with a sales team scenario. A workspace agent can:
- Consolidate information from call notes and customer research
- Qualify incoming leads
- Draft follow-up emails for sales managers
The company noted that this approach lets teams spend less time stitching together details and more time actually engaging with clients.
Workspace agents are currently available through research preview access on ChatGPT Business, Enterprise, and Edu plans, as well as ChatGPT for Teachers (K–12).
Privacy Filter: Free Tool for Sensitive Data Protection
Alongside workspace agents, OpenAI released Privacy Filter — a free, open-source model that automatically strips sensitive information from text before it reaches ChatGPT. The tool replaces personal data with generic placeholders like [PRIVATE_PERSON] or [ACCOUNT_NUMBER].
The filter scans text for eight categories of personal information:
- Names
- Addresses
- Email addresses
- Phone numbers
- URLs
- Dates
- Account numbers
- Secrets such as passwords and API keys
Privacy Filter is hosted on Hugging Face and GitHub, has 1.5 billion parameters, and is free to download and modify. OpenAI reported the model achieved a 96% score on a standard test using the PII-Masking-300k dataset.
The tool is designed for scenarios where users need to share tax returns, medical records, work emails containing client names, or code snippets with API keys in chat conversations.
Broader Context
The launch of workspace agents and Privacy Filter comes as OpenAI continues to expand its product lineup. Earlier in April, Sam Altman's company released ChatGPT Images 2.0 — a "thinking" image generator positioned as an advanced model for solving complex visual tasks.
Frequently Asked Questions
What are OpenAI workspace agents in ChatGPT?
Workspace agents are cloud-based AI assistants powered by Codex that teams can create once and share across an organization. They operate autonomously in the cloud, continuing tasks even when users are offline, and integrate with ChatGPT and Slack.
Who can access OpenAI workspace agents?
The feature is currently available through research preview access for ChatGPT Business, Enterprise, and Edu plan subscribers, as well as ChatGPT for Teachers (K–12). There is no general public rollout yet.
What is OpenAI's Privacy Filter?
Privacy Filter is a free, open-source model with 1.5 billion parameters that automatically redacts sensitive information from text before it is sent to ChatGPT. It is available on Hugging Face and GitHub for anyone to download and modify.
How accurate is OpenAI Privacy Filter?
According to OpenAI, the Privacy Filter achieved a 96% score on a standard test using the PII-Masking-300k dataset. It detects eight categories of personal information including names, addresses, emails, phone numbers, URLs, dates, account numbers, and secrets like passwords and API keys.
Read also
Google Enhances Opal AI Platform with New Autonomous Agents
Google has upgraded its visual AI workflow builder Opal with agent functionality that automatically analyzes tasks and selects appropriate tools for completion.
OpenAI Secures Record $110 Billion Round at $730 Billion Valuation
OpenAI closed the largest startup funding round in history at $110 billion, backed by Amazon, SoftBank, and Nvidia, with a $730 billion valuation.
AI Audit Uncovers Critical Liveness Bug in Ethereum's Nethermind Client
Octane Security's AI discovered a high-severity vulnerability in the Nethermind execution client that could have halted block production for 38% of Ethereum mainnet validators. The Ethereum Foundation awarded a maximum $50,000 bounty.
How ERC-8004 and x402 Standards Are Turning AI Agents Into Market Participants
Blockchain standards ERC-8004 and x402 are building the infrastructure for autonomous machine payments, challenging Big Tech's closed ecosystems and the attention economy model.
AI Agent Transaction Volume Is 15x Lower Than Bloomberg's Estimates, Says a16z Partner
a16z partner Noah Levine challenged Bloomberg's AI agent payment data, showing actual on-chain volume at $1.6–3M versus the reported $24M.
Weekly Recap: Bitcoin Tests $74K, Miners Dump Holdings, ChatGPT Boycott Grows
Bitcoin briefly touched $74,000 before retreating to $67,500. Public miners sold over 15,000 BTC in five months, traders flocked to Hyperliquid for oil and gold futures, and a ChatGPT boycott gained major traction.
