AI product notes: week of 29 June 2026

Smaller product updates do not always need a full article, but they can still matter if you own the platform, run the admin console, or build against the API.

Updates

• Anthropic: Claude memory from chat history now available to all users including free tier; Claude for Excel/PowerPoint add-ins gain cross-app context sharing (5 July 2026). Memory for free users is a significant consumer product parity move, while the cross-app Office add-in context sharing is directly useful for enterprise teams using Claude inside Microsoft 365 workflows.

• OpenAI: ChatGPT Business and Enterprise get a unified admin plugin management console for Codex (4 July 2026). Centralised plugin governance is a compliance prerequisite for many enterprises deploying Codex at scale, especially as the July 6 agent billing start increases scrutiny of what plugins are running.

• Anthropic: Elevated errors on Claude Opus 4.8 — active incident as of July 2 (2 July 2026). An active Opus 4.8 degradation immediately after the Fable 5 restoration and Sonnet 5 launch is operationally significant for any team that depends on Opus as their primary API model for agentic workflows.

• OpenAI: ChatGPT Business gets a new plugin management hub in Workspace settings (1 July 2026). Centralised plugin governance is a prerequisite for safe enterprise Codex rollouts; without it admins had no visibility into which plugins their workforce was connecting to sensitive data sources.

• Anthropic: Claude Code adds Trusted Devices requirement for Remote Control admin sessions (1 July 2026). Device verification closes a meaningful gap in the remote-control security model at a time when Claude Code agentic sessions are increasingly being run unattended on developer machines.

• Anthropic: Claude Code v2.1.197: Sonnet 5 as default, sandbox credential blocking, and org model restrictions (1 July 2026). Sandbox credential blocking and org-level model restrictions are meaningful security and governance improvements for enterprise teams running Claude Code at scale, especially given the new gateway rollout.

• Anthropic: Claude adds Trusted Devices for Remote Control Admins on Team and Enterprise plans (30 June 2026). Device verification closes a session-hijacking risk for enterprises that allow remote Claude Code control, especially relevant as agentic coding sessions handle production credentials.

• Anthropic: Claude Code adds org default models, 37% lower streaming CPU usage, and session readability improvements (30 June 2026). The org default model setting is a meaningful governance addition for enterprise admins managing Claude Code at scale, and the CPU/memory improvements matter for long autonomous coding sessions.

• Microsoft: Microsoft 365 Copilot June 2026 roundup: Copilot Notebooks expands to Copilot Chat users, Vision in Teams rolling out in July (30 June 2026). Notebooks reaching Copilot Chat users significantly expands who can use structured project context without a full Copilot license, lowering the adoption barrier for mixed-license organisations.