Here are the vendor updates that matter for UK teams this week. For each, we highlight what changed, why it matters operationally, and a simple action to consider.
OpenAI
Sora 2: storyboard editing and longer clips
- OpenAI added Storyboards in Sora 2 on web (beta) and increased video durations: all users up to 15s; Pro can generate 25s via storyboard. The Sora app rollout remains U.S./Canada for now. help.openai.com
What it means for UK teams: if you’re evaluating AI video for marketing or training, you can now test more realistic sequences and timing on web; but app‑based features may not be available in the UK yet.
Action to consider: run a 60‑minute storyboard test with 3 existing brand clips and your brand guidelines; log realism, timing, licence checks and watermarking settings. If policy is a blocker, see our AI Policy Pack (templates).
Anthropic (Claude)
Agent Skills for custom workflows
- New “Agent Skills” let Claude load organisation‑specific instructions, resources and optional code to improve specialised tasks (e.g., Excel, brand rules). Admin controls are available; Skills require the Code Execution tool. anthropic.com
What it means for UK teams: you can package repeatable workflows (tone of voice, spreadsheet formats, PDF templates) and reduce prompt wrangling.
Action to consider: start with one high‑volume task (board paper summaries, tender responses). Ship a single Skill, measure cycle time and rework.
Claude Code on the web
- Claude Code now runs coding tasks in Anthropic‑managed sandboxes via a browser, supporting multiple parallel jobs with PRs and change summaries. Research preview for Pro/Max users. anthropic.com
What it means for UK teams: useful for dev backlogs and routine fixes without local setup. Check data access rules and repository scoping.
Action to consider: trial on low‑risk repos with branch protections; track time‑to‑PR and bug escape rate versus your baseline.
Claude for Life Sciences
- Sector package adds connectors (Benchling, PubMed, BioRender, Wiley Scholar Gateway, 10x Genomics) and Skills for scientific workflows. anthropic.com
What it means for UK health charities/biotech: lower friction to literature reviews and protocol drafting; verify licensing and information governance.
Action to consider: run a compliant pilot with de‑identified datasets; add clinical safety review before scaling.
Google (Gemini)
Veo 3.1 preview and Maps grounding GA
- Veo 3.1/3.1 Fast (preview) adds video extension, multi‑image references, and first/last frame controls. ai.google.dev
- Grounding with Google Maps is now GA in the Gemini API (place/route awareness for responses). ai.google.dev
- Gemini API Additional Terms updated effective 17 Oct 2025 (includes age requirement 18+). ai.google.dev
What it means for UK teams: better control for product and location storytelling; ensure legal teams review new API terms before production use.
Action to consider: pilot a “store‑finder + local itinerary” or logistics prototype using Maps grounding; document data sources and opt‑outs.
Microsoft (Copilot)
October Copilot Chat roadmap highlights
- Microsoft’s October roadmap spotlights personalisation: “memory based on prompts” (launched) and in‑development features like temporary chat mode and memory citations in responses. microsoft.com
What it means for UK teams: memory features can boost continuity across chats—ensure your data retention policy and DPIA cover them.
Action to consider: ask IT to confirm tenant‑level settings and audit trails for Copilot memory before enabling organisation‑wide.
xAI (Grok)
Server‑side Tools GA
- xAI made Tools generally available (web_search, x_search, code_execution) for agentic use via API. docs.x.ai
What it means for UK teams: easier to prototype agents that browse or execute code server‑side—review guardrails and network egress.
Action to consider: restrict tool scopes and log tool invocations; start with read‑only browsing tasks.
Perplexity
Weekly ship (17 Oct): language learning, shopping/travel filters, finance feeds
- Perplexity shipped iOS language learning, adaptive filters for Shopping/Travel, and live earnings call audio plus insider activity in Finance. perplexity.ai
What it means for UK teams: a stronger research front‑end for consumer journeys and equity/market tracking; check source attribution meets compliance needs.
Action to consider: trial adaptive filters for product discovery pages; measure conversion uplift from summarised options.
Databricks
- Data Classification public preview (consolidated results + auto‑tagging UI). docs.databricks.com
- Asset Bundles in the workspace GA (collaborative edit/commit/test/deploy from UI). docs.databricks.com
- Multimodal model serving (vision) available across Foundation Model APIs. docs.databricks.com
What it means for UK teams: simpler controls for sensitive data and more predictable CI/CD for AI apps; vision workloads easier to host.
Action to consider: enable auto‑classification on a non‑prod catalogue; confirm tags drive downstream masking and audit in Unity Catalog.
Snowflake
- Release 9.32 (Oct 13–15) includes Notebooks replication GA and Auto‑refresh improvements for Iceberg; Table optimisation for Snowflake‑managed Iceberg is GA with billing from 20 Oct. docs.snowflake.com
- AI_EXTRACT AISQL function GA (Oct 16); Performance Explorer in preview (Oct 20); Clean Rooms updates landed today (Oct 23). docs.snowflake.com
What it means for UK teams: replication reduces DR gaps for Notebook‑based assets; Iceberg optimisation now billable—watch credits. New AI_EXTRACT and performance tooling can cut manual wrangling.
Action to consider: tag Iceberg optimisation spend; set alerts on credits; try AI_EXTRACT on a redacted sample to see if it replaces brittle regex rules.
Meta
- New parental controls for teens’ AI interactions across Instagram/FB: parents can disable 1:1 AI chats and see usage insights. about.fb.com
What it means for UK charities/education: clearer controls for safeguarding when using Meta AI in youth programmes; align with your safeguarding policy.
Action to consider: update your social media guidance and ensure staff know how to apply the controls on organisational accounts.
Quick Monday checks for leaders
- Budget: if you use Snowflake Iceberg, add a spend guardrail for table optimisation from 20 Oct.
- Policy: confirm your stance on AI memory features (Microsoft/Meta) and likeness controls (OpenAI Sora 2).
- Security: for Claude Agent Skills and Grok Tools, review code execution/network egress and restrict to approved domains.
If you’re moving from pilot to live service, this 30‑day cutover plan helps avoid surprises. Read the cutover plan. For content teams, consider our 14‑day AI content sprint.
What to watch next week
- OpenAI is reportedly preparing an AI‑powered browser (“ChatGPT Atlas”). If confirmed, expect a wave of “agent‑in‑the‑browser” workflows competing with Perplexity Comet and Edge Copilot. theverge.com
- Snowflake feature updates continue to roll weekly; watch for further Cortex and Clean Rooms changes. docs.snowflake.com