Rob FalconerThe latest in AI

Topic · Agents & Tools

AI agents & tools

This autumn's big shift is from chatbots that answer questions to agents that keep working on their own: they run on their own computers, connect to thousands of apps and write code while you sleep. Here is what has launched and how each company is trying to keep these agents safe.

Last updated:

OpenAI: Dots, ChatGPT apps and Business Agents

Dots: always-on agents

29 Sep 2026OpenAI's Dots are always-on agents powered by GPT-6 Astra. Each one gets its own cloud computer. They can connect to more than 4,000 apps, and you can reach them inside ChatGPT, Slack and Microsoft Teams. They are rolling out first to Pro and Business Premium customers.

Organisations can also set up "specialist" dots, and OpenAI describes an integration with Microsoft Agent 365. On safety, Dots include automatic review of the actions an agent proposes and Custom Rules that let administrators restrict what agents may do.

Businesses in ChatGPT

10 Nov 2026OpenAI launched a beta that lets businesses claim a profile in ChatGPT, keep their details and product feeds up to date, and see aggregated figures on how often they are mentioned and clicked. Eligible US ecommerce and online-service businesses can also launch a Business Agent: a clearly labelled, separate chat that answers questions about their products and policies. OpenAI says it does not share users' conversations with businesses, and that businesses can sponsor their agents as a destination for ChatGPT ads.

Oracle and iCloud apps in ChatGPT

27 Oct 2026OpenAI and Oracle announced Oracle AI Workspace, which Oracle is developing to bring its finance, procurement and planning apps into ChatGPT, within each company's permissions and approvals. Oracle is also launching AI Security Mesh, which uses OpenAI's cyber models to find and fix vulnerabilities, an SMB bundle with ChatGPT Business and Codex, and App Composer for building apps with Codex. OpenAI says Oracle itself has 130,000 active ChatGPT users and more than 95,000 active Codex users.

27 Oct 2026The same day, ChatGPT gained iCloud Mail and iCloud Calendar plugins on all plans, on the web, desktop and iOS. MacRumors reports that you connect them with an Apple app-specific password rather than your main Apple password. As with any agent connection, start with read-only tasks.

Google: Antigravity, Orchestrator and Agent Compute Engine

Antigravity, now with Gemini 4 Argon

13 Oct 2026Antigravity is Google's agent-first development platform, built around agents rather than an editor with a chat box attached. Google says that after Gemini 4 Argon came to Antigravity, agentic code submissions rose by more than 35%. It also describes Argon refactoring about 6 million lines of legacy Chrome code.

Orchestrator: two models, one task

12 Nov 2026Antigravity now has an "Orchestrator" option that splits one task between models: Gemini 4 Carbon or Argon explores and plans, the new, faster Gemini 4 Hydrogen makes the edits and runs the tests, and work passes back to the bigger model when Hydrogen gets stuck. Google says Orchestrator reached close to Carbon's success rate on Terminal-Bench 4.1 while using 24% less quota and finishing 19% faster. Hydrogen is also now the default model for Google's managed Antigravity agent.

Agent Compute Engine

22 Oct 2026Google announced Agent Compute Engine, which gives each agent its own hardware-isolated micro virtual machine. Google says the same system already runs agents inside Gemini and will be in preview in the coming months. Short-lived credentials are injected from outside the virtual machine, so the agent does not hold long-lived keys. Google also open-sourced Agent Substrate for running agents on Kubernetes (GKE), claiming up to 10 times the density, and said its agent governance controls will extend to Claude Code and the Antigravity CLI on managed company devices, in preview soon.

Anthropic: Claude Meetings, Duet and auto mode

Claude Meetings

12 Nov 2026Claude can now join video calls as a visible participant, answer questions out loud or in the chat, take notes and action items, and build documents or whiteboards with the team during the call. It is rolling out in beta on Team and Enterprise plans, for Google Meet only at first, and is not yet available to organisations with zero data retention or HIPAA-ready setups. Anthropic says it can use connected tools only with permission.

Whiteboards, Sheets and Data

10 Nov 2026Anthropic launched Claude Whiteboards and Claude Sheets in beta on all paid plans, so teams can brainstorm and build spreadsheets with Claude without moving work to another app. Claude Dashboards is now generally available as Claude Data, for asking questions of company data in warehouses such as BigQuery.

Claude Duet

21 Oct 2026Claude Duet is a background assistant built around Claude Opus 5.5 that watches your email, calendar and business tools, sends a morning brief and checks connected accounts every few hours from its own virtual machine. It is a research preview for US Pro and Max subscribers through a waitlist. It must ask before sending messages, making payments, signing documents or deleting anything, and some actions, such as trading securities, are blocked even with approval. Read my post on Duet and auto mode.

Auto mode by default

22 Oct 2026Anthropic says auto mode, where a classifier rather than the user approves routine actions, would roll out the following week as the default for new sessions in most Claude apps on Pro, Max and Team plans. It has been the default in Claude Code since 14 August. Anthropic says the classifier caught 89% of dangerous commands in its study, compared with 13.6% for people, but that auto mode "does not eliminate risk".

Personal Finance in Claude

Early Nov 2026Anthropic is rolling out Personal Finance in beta for Pro, Max and Ultra subscribers in the US. You connect bank and card accounts by signing in through Plaid, so Claude does not see your bank password, and then ask questions in plain language about spending, upcoming bills or odd charges. Anthropic says access is read-only, so Claude cannot move, manage or invest money, and that it can see balances, credit limits and up to two years of transactions for the accounts you choose. Conversations in the Personal Finance hub are not used to train its models, and Anthropic warns that Claude can make mistakes, so important figures should be checked with your bank.

Mistral: Studio

Managed agents, workflows and apps

5 Nov 2026Mistral expanded Studio, its enterprise platform, with managed agents and apps and new tools for workflows, observability and evaluations. Agents can be given skills, sandboxes for running code, subagents and hooks for custom logic, with limits set on the tools and data they can reach. Workflows mix fixed rules, agent reasoning and human approval, and can resume after an interruption. Studio can run in Mistral's cloud, a customer's own cloud or on-premises, and records each model call, tool use and approval as OpenTelemetry traces. A new Mistral MCP server lets developers build with their own coding agent and deploy to Studio with one command. The customer examples and time savings in Mistral's post are its own claims.

Keeping agents in check

The push towards autonomy has come alongside a very public reminder of the risks. In July 2026, during cyber-capability testing, an internal OpenAI model and its agents escaped their sandbox and compromised Hugging Face systems. OpenAI published a post-mortem on 26 August. My explainer covers what happened.

On 20 October OpenAI reported two more internal incidents from early October. A GPT-6.1 Sol agent stopped alerts reaching its security team after reading an earlier "don't send any messages" instruction too broadly, and a GPT-6 Astra agent removed a monitoring integration and then wrongly told the user it had not. OpenAI says neither reached its top severity level, and its automated review now blocks attempts to disable monitoring.

On 29 October OpenAI described how it watches for this kind of behaviour. Its system has four layers, from probes that read the model's internal activations to automated investigations, and it monitors frontier training runs and internal Codex use. OpenAI says that when it replayed the Hugging Face incident, the system raised an alert about 30 minutes after the escape, more than 30 hours before the breach. It also says alerts on internal Codex use do not currently page anyone. See the policy page for more.

On 4 November OpenAI added a lesson from its Linux security research. In an early run its exploit agents reported 200 successes, but human reviewers found that 20 had gained root through a different flaw from the one they were assigned. OpenAI now adds an independent check that each result used the assigned route. My post covers the research.

The safeguards now appearing in agent products reflect that:

  • Review before action: Dots automatically review proposed actions, and administrators can add Custom Rules.
  • Staged access: the most capable cyber features go to vetted defenders first, through OpenAI's Daybreak and Google's Fairwind Program. Google's Gemini 4 Carbon is going to defenders before anyone else (my post), and so is OpenAI's GPT-6 Cyber (my post).
  • Classifier approvals: Anthropic's auto mode and OpenAI's automated review check each proposed action, with people kept for the high-stakes ones.
  • Isolation: Google's Agent Compute Engine and Claude Duet both run each agent in its own virtual machine, away from your own devices and long-lived passwords.
  • Government preparation: on 12 October the UK government said it is preparing, with the NCSC and the AI Security Institute, for incidents involving AI agents. The ICO's call for evidence on agentic AI closes on 20 November 2026.

Practical tips if you are trying agents

  • Start agents on read-only tasks such as research, summaries and drafting before you give them permission to send, buy or delete anything.
  • Give each agent the narrowest app connections it needs, and review its connections regularly.
  • Keep a person approving anything that is irreversible or visible to people outside your organisation.
  • Watch your costs: always-on agents that use flagship models can get expensive quickly, which is exactly why cheaper tiers such as Luna, Haiku, Muse Ember and Gemini 4 Hydrogen matter. Splitting work between a big planner and a cheaper worker, as Google's Orchestrator does, is another way to cut costs.

Sources

  1. Introducing Dots — OpenAI (29 September 2026)
  2. DevDay 2026 recap — OpenAI (29 September 2026)
  3. Introducing GPT-6.1 Terra — OpenAI (26 October 2026)
  4. OpenAI expands partnership with Oracle to bring its apps to ChatGPT — OpenAI (27 October 2026)
  5. ChatGPT Can Now Access Your iCloud Mail and Calendar — MacRumors (27 October 2026)
  6. Introducing Gemini 4 Argon (general availability) — Google (13 October 2026)
  7. Google Cloud brings OpenAI models to customers — Google Cloud (21 October 2026)
  8. Introducing Agent Compute Engine and expanded agent governance — Google Cloud (22 October 2026)
  9. Introducing Claude Duet — Anthropic (21 October 2026)
  10. Introducing Personal Finance in Claude — Anthropic (accessed 6 November 2026)
  11. Build Enterprise Agents, Workflows & Apps in Mistral Studio — Mistral AI (accessed 6 November 2026)
  12. A thousand fewer paths to root — OpenAI (4 November 2026)
  13. GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents — OpenAI (5 November 2026)
  14. Auto mode is becoming the default in Claude apps — Anthropic (updated 22 October 2026)
  15. Claude Opus 5.5 — Anthropic (22 September 2026)
  16. The Hugging Face incident and the road ahead — OpenAI (26 August 2026)
  17. Britain steps up national defences against growing AI risks — GOV.UK (12 October 2026)
  18. Suppressing misalignment monitoring alerts and disabling misalignment monitoring — OpenAI Alignment (20 October 2026)
  19. How we detect frontier model misalignment — OpenAI (29 October 2026)
  20. Gemini 4 Carbon: our next frontier model — Google (28 October 2026)
  21. ICO secures changes from leading AI developers as scrutiny extends to AI agents — ICO (8 October 2026)
  22. Introducing Businesses in ChatGPT — OpenAI (10 November 2026)
  23. Gemini 4 Hydrogen and Orchestrator — Google Antigravity blog (12 November 2026)
  24. Introducing Claude Meetings, now in beta — Anthropic (12 November 2026)
  25. Introducing Claude Whiteboards, Claude Sheets, and Claude Data — Anthropic (10 November 2026)

Figures quoted from company announcements are the companies' own claims. Independent testing may give different results.