AI product notes: week of 24 August 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 Code fixes bash auto-approval arithmetic bug and ultrareview early-exit on cloud session failure (30 August 2026). The bash arithmetic auto-approval bug is a security fix, any agentic pipeline running Claude Code in auto mode should update immediately to ensure dangerous shell commands with arithmetic expressions are no longer silently approved.

• Microsoft: Copilot in 30: Microsoft launches a partner-led M365 Copilot Business trial programme for SMBs through December 2026 (30 August 2026). SMBs buying Microsoft 365 Copilot Business through a CSP partner can now enter a structured 30-day guided trial programme rather than purchasing cold, giving IT decision-makers a lower-risk on-ramp through the end of 2026.

• Google: Gemini Notebook adds interactive reports with embedded Studio content (30 August 2026). Gemini Notebook can now output shareable interactive reports, not just text or audio, making it more useful for presenting research findings to stakeholders without leaving the Notebook environment.

• Google: Gemini ‘Take notes for me’ controls coming to Meet Hardware touch controllers on 31 August 2026 (29 August 2026). Conference room participants can now control Gemini note-taking from the room’s touch panel like any other meeting control, so note-taking no longer requires someone to have their laptop open in Companion mode.

• Google: Gemini Enterprise agents now emit OpenTelemetry-aligned telemetry in experimental preview (29 August 2026). Teams already running OpenTelemetry for their services can now route Gemini agent spans into the same dashboards and alerting pipelines without a custom adapter, making AI workflow debugging a first-class citizen in standard observability tooling.

• Anthropic: Anthropic adds personal keys and service-account keys to the Claude Console (28 August 2026). Admins can now issue trackable, least-privilege API credentials for both humans and automated services, and revoking an employee’s access automatically kills their personal key, closing a common audit gap in shared-key setups.

• OpenAI: ChatGPT adds multi-account Google support: Gmail, Calendar, and Contacts from personal and work accounts in one conversation (28 August 2026). Users who keep separate work and personal Google accounts can now ask ChatGPT to find a free meeting slot or search across both inboxes at once, removing the need to switch sessions.

• Anthropic: Claude Code v2.1.250 ships bug fixes and reliability improvements only (28 August 2026). Long-running Claude Code sessions that use extended thinking were silently losing cache context once an hour; this fix prevents unexpected token cost spikes and context drops in agentic workflows.

• OpenAI: ChatGPT temporary chat now supports personalised responses with memory, plugins, and custom instructions (27 August 2026). Power users who rely on temporary chat for private queries can now get personalised, context-aware responses without sacrificing the no-history default, a meaningful middle ground between full chat history and a completely blank session.

• Anthropic: Claude Code adds self-generated feedback reports and cross-session messaging changes in latest release (27 August 2026). Claude Code can now self-diagnose and document its own failures during long agentic runs, giving developers a structured paper trail when something goes wrong mid-workflow rather than having to reconstruct the failure manually.

• Google: CodeMender on Gemini Enterprise Agent Platform adds Gemini 3.6 Flash and 3.7 Flash support plus unrestricted verification mode (27 August 2026). Security teams using CodeMender can now run vulnerability scans backed by Google’s newest and most capable Flash models, and the unrestricted verification flag makes it practical to confirm exploits in sandboxed environments without wrestling with command-policy blocks.

• Google: Gemini app now generates interactive simulations and 3D models in chat responses (27 August 2026). Gemini in the app can now build a working 3D model or simulation as its answer, turning conceptual questions about science, maths, or engineering into hands-on interactive tools rather than text explanations.

• Anthropic: Claude API Skill open-sourced: pre-built Agent Skill for the Messages API across 8 languages, bundled with Claude Code (25 August 2026). Developers can now drop a maintained, always-current API reference directly into their Claude Code or agent workflow instead of hand-crafting documentation context.

• Anthropic: Claude Code v2.1.245 fixes startup crash on glibc 2.44 Linux distributions (25 August 2026). Claude Code fails to start on Arch Linux, CachyOS, and Fedora Rawhide with the latest glibc, upgrade to v2.1.245 to fix it.

• Anthropic: Claude Managed Agents web_search and web_fetch can now be restricted to an allowlist of domains (25 August 2026). Enterprise teams can now lock Claude agents to an approved list of external domains, reducing the risk of sensitive queries reaching unintended third-party sites.

• Anthropic: Claude Rate Limits API is now available: admins can query org and workspace rate limits programmatically (25 August 2026). Teams running Claude at scale can now pull rate-limit data into their own dashboards and alerting pipelines instead of checking the Console by hand.

• Google: Gemini ‘Take notes for me’ controls arrive on Google Meet Hardware touch controllers from 31 August 2026 (25 August 2026). Conference-room teams can now control Gemini note-taking from the room’s own touchscreen, pausing for off-the-record moments without anyone needing a personal laptop in Companion mode.

• Google: Google Workspace data import advanced mode is now GA, supporting large-scale OneDrive-to-Drive migrations (25 August 2026). IT teams moving large file estates from OneDrive to Drive can now run the migration in bulk with permissions intact and no additional licensing cost.

• Anthropic: Memory for Claude Managed Agents enters public beta under the standard managed-agents header (25 August 2026). Developers building production agents can now add persistent memory without juggling an extra beta flag, lowering the barrier to multi-session context in API-driven workflows.

• Anthropic: Claude Code reduces claude-api skill context cost by 87% and adds mid-turn slash command access (24 August 2026). Claude Code sessions that use the built-in API skill will consume dramatically less context window per turn, leaving more room for code and reasoning, an immediate win for large-codebase workflows.

• Google: Gemini Notebook adds a full notebook copy capability in Workspace (24 August 2026). Teams that reuse research notebook structures, for recurring reports, audit templates, or training materials, can now copy a complete notebook rather than rebuilding sources and studio items from scratch each time.