AI expands accessibility scope

Artificial intelligence is moving from optional classroom software to shared campus infrastructure, and that is widening the list of places where accessibility can fail. (openai.com) OpenAI said on April 22 that “workspace agents” are rolling out to ChatGPT Business and Enterprise, with release notes saying the feature is arriving over the next few weeks and can run inside ChatGPT or Slack. OpenAI describes the agents as shared, Codex-powered tools that automate repeatable workflows across connected apps. (openai.com) OpenAI’s “company knowledge” feature is already available for ChatGPT Business, Enterprise, and Edu, pulling organization-specific context from connected apps so users can get answers without leaving a chat. OpenAI says the feature includes citations, admin controls, and security and privacy settings for workplace use. (openai.com) Google has been making the same shift in schools: in June 2025, it expanded Gemini in Classroom to users 18 and older across all Google Workspace for Education editions, after first limiting it to paid add-ons. Google said the rollout brought more than 30 artificial-intelligence tools into Classroom and added an admin setting to control access. (workspaceupdates.googleblog.com) Anthropic entered the higher-education market on April 2, 2025 with Claude for Education, a version of Claude aimed at teaching, learning, and administration. Anthropic said its “Learning mode” is designed to ask questions and guide students through problems instead of simply giving answers. (anthropic.com) As these systems move into email, file storage, chat, assignments, and student support, accessibility stops being a question about one chatbot window and becomes a question about entire workflows. Google’s admin documentation says Gemini in Classroom is on by default for users marked 18 and older, while OpenAI says workspace agents operate within organizational permissions and controls. (knowledge.workspace.google.com) (openai.com) That changes what schools and employers have to evaluate. A screen reader user, a student relying on captions, or an employee using keyboard navigation may be affected not only by model output, but by how an agent pulls from Slack, SharePoint, Google Drive, or other connected systems and returns the result. (openai.com) School systems are starting to answer that with rules, not bans. New York City Public Schools’ March 2026 guidance allows artificial intelligence for tasks like lesson planning and drafting communications, but bars it from grading, discipline, and individualized education programs, using a red-yellow-green framework for staff. (govtech.com) The argument over schools is no longer just whether students should use chatbots for homework. It is whether institutions can govern tools that are being built into everyday teaching, advising, and office work faster than those tools become normal. (newyorker.com)