Here’s a crisp roundup of the most useful AI announcements in the last 7–10 days for UK SMEs, charities and chambers: what changed, why it matters, and the action worth taking.
OpenAI
-
GPT‑5.1 options arrive in ChatGPT for Enterprise/Edu, with Instant and Thinking modes, routed by GPT‑5.1 Auto; admins must enable access. help.openai.com
What it means for UK teams: You can balance speed vs depth per use case (quick chat vs heavier analysis) while keeping central control.
Action to consider: Ask IT to enable GPT‑5.1 for a pilot group and document when to use Instant vs Thinking in your playbooks. -
Apps and an Apps SDK are rolling out to Enterprise/Edu, with partner apps such as Canva and Figma, plus new synced connectors (Azure Boards, Basecamp, Zoho CRM). help.openai.com
What it means for UK teams: Fewer tab‑swaps; more tasks completed inside one governed interface.
Action to consider: Prioritise 2–3 high‑value apps and switch them on behind RBAC; agree data‑handling boundaries before wider rollout. -
New MCP access connectors (e.g., Stripe, Semrush, Vercel, Monday.com) were added for Business/Enterprise/Edu/Pro/Plus where connectors are enabled. help.openai.com
What it means for UK teams: Easier secure data pull from key SaaS, reducing copy‑paste risk.
Action to consider: Map which line‑of‑business tools you’ll connect first; capture connector scope in your DPIA and audit logs. -
ChatGPT Atlas (the new desktop browser) received updates on 18 and 25 Nov—vertical tabs, memory‑aware responses, and a “safe search” toggle. help.openai.com
What it means for UK teams: Better research ergonomics for analysts; ensure safety settings align with policy.
Action to consider: Standardise Atlas settings via an onboarding checklist; train staff to cite sources when using AI‑assisted research.
Anthropic (Claude)
-
Claude Opus 4.5 launched on 24 Nov with stronger coding, agentic workflows, and office tasks (slides, spreadsheets). Pricing is listed at $5 input / $25 output per million tokens; caps eased for Opus users in apps and Claude Code. anthropic.com
What it means for UK teams: If you postponed heavy coding or spreadsheet automation due to cost or limits, Opus 4.5 may now be viable.
Action to consider: Run a side‑by‑side 1‑week pilot against your current model for a typical backlog (bug fixes, refactors, Excel models), tracking task completion rate and token spend. -
Anthropic’s page notes improved long‑running/agent tasks and an effort parameter to tune speed vs capability, plus increased usage limits for Max and Team Premium. anthropic.com
What it means for UK teams: You get predictable cost control by dialling effort up/down without re‑engineering your prompts.
Action to consider: For nightly data/report automations, try lower effort settings; reserve high effort for thornier analysis.
Google (Gemini)
-
Gemini 3 unveiled on 18 Nov with upgraded reasoning and multimodal capability; available across the Gemini app, AI Studio and Vertex AI. blog.google
What it means for UK teams: If you already use Google Workspace/Cloud, Gemini 3 may simplify procurement and governance.
Action to consider: Ask your Google partner to demo Gemini 3 in your own docs and slides; include it in any December vendor bake‑off. -
Google is also bringing Gemini 3 to Search (AI Mode) with dynamic, interactive layouts—first markets are rolling out now. blog.google
What it means for UK teams: Staff will encounter AI‑generated answers more often; your content strategy should focus on trustworthy, source‑rich assets.
Action to consider: Update your content QA to ensure citations and evidence are clear; see our playbook on trust‑first AI content.
xAI (Grok)
-
Grok 4.1 released 17 Nov; xAI says it’s preferred in blind human evals and available across web, X, iOS and Android. x.ai
What it means for UK teams: If your social team works heavily in X, Grok’s native integration may remove friction for trend and content drafting.
Action to consider: Trial Grok 4.1 for community management replies; measure moderation effort saved and factuality vs your current model. -
New Grok 4.1 Fast model and Agent Tools API add 2M‑token context and tool‑calling for agent use cases. x.ai
What it means for UK teams: Long documents and multi‑step tasks (support bookings, finance checks) can run inside xAI’s API.
Action to consider: If you build on xAI, start a proof‑of‑concept for a constrained task (e.g., booking changes) with human‑in‑the‑loop.
Microsoft
-
Security Copilot agents are now included for all Microsoft 365 E5 customers over the coming months; 12 new Microsoft‑built agents span Defender, Entra, Intune and Purview. Rollout has begun. microsoft.com
What it means for UK teams: If you already license E5, you’ll gain AI help for alert triage, identity risk and data protection without extra line‑item costs.
Action to consider: Ask IT when your tenant will be enabled and request a playbook for scope, audit logging and who can invoke which agents. -
Microsoft also highlighted broader Copilot and agent updates at Ignite, including Windows 365 for Agents (public preview) and a new SMB offer (“Microsoft 365 Copilot Business”) at $21 per user/month from December. microsoft.com
What it means for UK teams: A clearer entry price for sub‑300‑user organisations; assess value vs current Copilot licensing.
Action to consider: Run a 30‑day pilot in two departments and benchmark weekly time saved on email, meetings and document prep.
Snowflake
-
New SQL‑native AI_COMPLETE function reached General Availability on 21 Nov, enabling prompt‑based completions (text and images) directly in Snowflake. docs.snowflake.com
What it means for UK teams: You can prototype AI jobs inside governed SQL without shipping data to external services.
Action to consider: Ask your data team to pilot a simple classification or summarisation workflow using AI_COMPLETE and document approval steps. -
Cortex Agents are GA, with a managed MCP server to standardise agent access to Snowflake data. Integration for Microsoft Teams and Microsoft 365 Copilot is GA too. docs.snowflake.com
What it means for UK teams: A cleaner, standards‑based way to let AI agents query Snowflake with RBAC and OAuth.
Action to consider: Define which datasets are agent‑eligible and set RBAC upfront; pilot the Teams integration with a narrow question set.
Perplexity
-
Perplexity introduced assistant memory to personalise answers based on preferences and prior conversations. Published 26 Nov. perplexity.ai
What it means for UK teams: Researchers and support teams may get more consistent outputs across sessions—but confirm privacy settings first.
Action to consider: Enable memory for a small group with clear guardrails (what it may remember) and test quality uplift on recurring tasks. -
Comet for Android launched on 22 Nov, bringing the AI browser to mobile with voice and cross‑tab summarisation; Perplexity’s changelog details the release. perplexity.ai
What it means for UK teams: Field and frontline staff can research and act from phones with clearer task audit trails than standard browsing.
Action to consider: Trial Comet on COPE devices for policy‑approved research tasks; ensure mobile MDM policies are in place.
Quick risk and cost checks for this week
- Data protection: Any use of assistant “memory” or connectors should be captured in your Record of Processing and DPIA. Map retention defaults and opt‑out paths.
- Hidden costs: New features often default on. Add token caps and monthly spend alerts; see our AI cost guardrail.
- Procurement timing: If you’re mid‑renewal with Microsoft or Google, incorporate these changes into your negotiation pack. Our two‑week vendor bake‑off shows how to compare like‑for‑like.
What to watch next week
- Google timing on broader access to Gemini 3 “Deep Think” and additional regions for Search AI Mode. blog.google
- Microsoft tenant‑level timelines for enabling Security Copilot agents across E5 estates—look for admin centre notices. microsoft.com
- OpenAI connector catalogue growth and admin controls as MCP‑based integrations expand. help.openai.com
- Anthropic early customer results on Opus 4.5 effort tuning and agent reliability on long‑running tasks. anthropic.com