ChatGPT

By August 2026 OpenAI's usage data described an older and more visual user base: users 35 and over increased in over 90% of countries, multimedia conversations reached 7.8% globally and 4.6% in Japan after the April 2026 release of ChatGPT Images 2.0, and work-related artifact generation more than doubled globally. In the same month the official desktop app arrived on Linux on August 11, a preview carrying ChatGPT, ChatGPT Work, and Codex across Ubuntu, Debian, and Fedora, and a plugin linked Apple's Messages app inbox so users could organize, analyze, edit, send, and delete messages from the ChatGPT interface. ChatGPT for Teens followed on August 19 for users aged 13 to 17, with a learning mode and parental controls, and a confirmation screen for sensitive images. AI developer Simon Willison's September 2 teardown split ChatGPT Work, launched in July 2026, into a cloud version and a local desktop version, with the cloud version limited to paid plans of $20 or more per month.

September widened the product line and the attack surface. On September 11 OpenAI introduced Data agent for ChatGPT Work, connecting approved sources including Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB, and Snowflake and importing files from Google Drive or SharePoint; the company said NTT Data joined its alpha program. The same day, Check Point Research reported a covert channel linking the code execution containers of two separate ChatGPT accounts, demonstrated by retrieving data from a victim's connected Gmail while the victim saw only a normal reply. OpenAI confirmed the vulnerability is fixed. ChatGPT for Financial Services also arrived, pairing company-hosted financial data with the GPT-6 Astra model, designed with Morgan Stanley and Evercore and incorporating data from Daloopa, PitchBook, LSEG News, and Crunchbase, with access limited to eligible financial institutions by contacting OpenAI. On September 15, coverage described Project Lily, an internal project in which several hundred contract staff read real user conversations and evaluate response quality; usernames are hidden and a Privacy Filter replaces names, addresses, email addresses, and phone numbers with labels, but OpenAI says the filter may miss uncommon identifying information or details whose meaning depends on context. ChatGPT for individuals uses conversations for model improvement by default.