AI product notes: week of 28 September 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

• Microsoft: Copilot Chat gets a new unified interface in Teams and Outlook, with Cowork and Work IQ access from one panel (4 October 2026). Users who relied on the Copilot shortcut in the Teams chat list will find it gone; the replacement is the left-side nav panel, which now also surfaces Cowork and Work IQ, giving a single place to reach all Copilot capabilities without context switching.

• OpenAI: gpt-rosalind-research billing also starts October 5, 2026 (4 October 2026). Teams given early access to gpt-rosalind-research should verify their spending limits before today’s billing start, or requests will accrue charges against live budgets without further warning.

• Google: Google Vids upgrades its AI voiceover engine to Gemini 3.8 Flash Lite TTS (4 October 2026). Workspace users creating AI-narrated videos in Google Vids will get noticeably more natural voiceovers from the upgraded Gemini 3.8 Flash Lite TTS engine, at no extra cost as part of their existing Workspace subscription.

• Anthropic: Anthropic discloses that some Claude messages sent during the September 29 outage were permanently lost (2 October 2026). Users who relied on Claude conversations from 29 September should check their history for gaps, and API teams should audit any automated jobs that ran credit top-ups on 1 October, balances may have shown as insufficient even after payment.

• Anthropic: Claude Code v2.1.288 closes a bypassPermissions file-deletion hole and fixes 89 bugs (2 October 2026). The bypassPermissions deletion hole is a meaningful security fix for anyone running Claude Code in CI/CD pipelines or shared environments, update to v2.1.288 before your next automated run.

• Microsoft: Microsoft confirms Copilot Managed Runtime enters preview, expanding Agent 365 cost management to Code and Runtime (2 October 2026). Enterprise teams building apps through Copilot Cowork, Code, or Copilot Studio can now track their AI compute spend in one dashboard, and Copilot Managed Runtime means those apps run securely inside the organisation’s own M365 environment rather than on external infrastructure.

• Anthropic: Claude Code v2.1.287 ships Claude Mods: already covered, plus three security fixes (2 October 2026). Developers using Claude Code behind a Bedrock-backed gateway who set CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS should update immediately, the flag was silently failing to remove structured-output format from requests, causing gateway rejections.

• OpenAI: Codex 0.160.0 adds agent command-centre history browsing and projectless workspace sessions (2 October 2026). The new agent command-centre history browsing makes it easier to audit and resume past Codex agent runs, while projectless workspace sessions let developers spin up quick tasks without first setting up a project structure.

• Anthropic: Barclays commits to 50% developer adoption of Claude Code by end of 2026 in joint Anthropic announcement (1 October 2026). Barclays is the first major global bank to publicly commit to a specific Claude Code developer-adoption target, giving enterprise procurement teams a concrete reference point for how a regulated financial institution is scaling AI coding agents.

• Anthropic: Barclays will have 50% of its software developers on Claude Code by end of 2026 (1 October 2026). Barclays’ commitment to put half its software developers on Claude Code by year-end is a significant public benchmark for large-bank AI adoption that gives enterprise procurement teams a reference point.

• Microsoft: Copilot ‘Today’ personalised command centre enters private preview in October 2026 (1 October 2026). Today is the next step in Microsoft’s vision of a Copilot that proactively organises your workday rather than waiting for a prompt, admins interested in early access should register for the Frontier preview programme now.

• Microsoft: Realtime voice interactions in Word and PowerPoint Copilot delayed, re-releasing in October 2026 (1 October 2026). Tenants that saw realtime Copilot voice in Word and PowerPoint disappear should expect it to return with changes in October; admins who have not yet reviewed voice feature policies should do so before it re-appears.

• Google: Replicated voice creation broke in the Gemini API and AI Studio on 1 October: now resolved (1 October 2026). Any pipeline relying on the Gemini API to clone voices from audio samples was broken for several hours on 1 October; the service is now restored, but teams should verify their integrations and check for failed jobs.

• OpenAI: Codex CLI adds optional account-security setup reminders for local ChatGPT-signed-in sessions (1 October 2026). Teams managing local Codex installations should be aware of the new security reminder banner, which may prompt users to complete MFA or other account-hardening steps they skipped at sign-in.

• Google: Ask Gemini in Chat promotional usage limits expire October 1, 2026 (30 September 2026). Teams that have been relying on Ask Gemini in Chat heavily during the promotional period should check their usage and watch for the updated limit figures Google promised to publish in the Help Center.

• Anthropic: Claude for Government also launches Claude Code CLI and Claude for Microsoft 365 in early access (30 September 2026). Government development teams can now request early access to Claude Code’s CLI and a Claude add-in for Microsoft 365, both running inside the agency’s FedRAMP High boundary, without waiting for a separate commercial rollout.

• Anthropic: Anthropic launches AI-conducted user research study on Claude and Claude Code from September 29 to October 6 (30 September 2026). Claude and Claude Code users on any paid or free plan can participate in the study this week and optionally make their session public, giving Anthropic direct input on product direction.

• OpenAI: Codex 0.159.3 adds optional account security setup reminders for local ChatGPT-signed-in sessions (30 September 2026). The security prompt nudge helps developers who set up Codex locally with ChatGPT sign-in to close any gaps in their account security before running autonomous coding tasks.

• Anthropic: Anthropic reports ~40-minute partial outage across Claude.ai, Claude Code, Claude Cowork, and Claude API on 29 September (29 September 2026). The outage briefly broke API calls, new conversations, and file uploads for users across Claude.ai and Claude Code, enterprise teams relying on Claude for production workloads should review their incident-response runbooks.

• Anthropic: Anthropic runs ‘What do you want from AI?’ public study using Anthropic Interviewer through 6 October (29 September 2026). Claude and Claude Code users on any paid or free plan can participate in shaping Anthropic’s understanding of what people actually want from AI, interviews are open until 6 October 2026.

• OpenAI: ChatGPT Ads expands to Southeast Asia and Taiwan (29 September 2026). Advertisers targeting Southeast Asian and Taiwanese audiences can now reach ChatGPT’s free-tier users in those markets, while publishers and brands in the region should prepare for the ad-supported model arriving on the platform they may already use or compete against.

• OpenAI: Codex Sep 28 release: MCP OAuth client secrets, WebSocket bearer tokens, transparent-background image generation, and TUI copy-paste (28 September 2026). Developers connecting Codex to MCP servers that require OAuth client secrets can now do so natively, and all elevated terminal commands now default to requiring approval, a meaningful security tightening for agentic workflows.

• Google: Google Workspace meeting resources get a centralised hub on the new Google Meet homepage (28 September 2026). Workspace admins and users can expect a reorganised Meet homepage from late September, consolidating meeting prep, recordings, and AI-assisted resources into one place.