Analysis
AI news and analysis
Latest posts
Anthropic's private beta lets a company run the whole Claude Managed Agents service in its own cloud or data centre. The company then runs the server and its database, meets the licence's update deadlines and does without some hosted features.
On 12 November 2026 Anthropic started rolling out Claude Meetings in beta on Team and Enterprise plans. Claude joins Google Meet calls, answers questions, builds documents and keeps a transcript. Team plans have it on by default, and Enterprise plans start with it off.
On 10 November 2026 OpenAI opened Businesses in ChatGPT in beta. Eligible businesses can claim their profile in Business Hub, correct their details, add product feeds and see aggregated insights. Business Agents, which answer customers inside ChatGPT, start with US ecommerce and online service businesses.
Google opened Gemini 4 Carbon to paying developers and enterprises on 10 November 2026. It costs $2 per million input tokens and $10 per million output tokens until 30 November, then $4 and $20. Gemini 4 Argon's API price rises on the same date, and Carbon spends more tokens per turn than Argon.
On 9 November 2026 OpenAI put ChatGPT Enterprise on AWS Marketplace. It is billed through AWS, and eligible spend can count toward an AWS commitment. OpenAI stays the data processor, and pricing is by private offer only.
Anthropic released Claude Fable 5.2 on 9 November 2026. It costs $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1. Four changes can break code that moves from 5.1, and the model needs 30-day data retention unless a customer qualifies for an exemption.
On 5 November 2026 Microsoft said Claude models in Copilot will process data inside the EU Data Boundary by the end of 2026. Today Claude sits outside it, and it is off by default for organisations in the EU, EFTA and the UK.
On 5 November 2026 Anthropic added a security center, a parent admin console and new access controls to Claude Enterprise. Its article also sets out an EDR integration for Cowork. Inference hooks are now generally available. The rest is in beta, and each feature has a limit a security review is likely to ask about.
On 3 November 2026 the UK government set out four planned measures for AI agents that operators cannot monitor or control. They would come under the Cyber Security and Resilience Bill. A second factsheet describes a power to direct essential services to stop using a supplier's product, with fines of up to 10% of turnover.
On 30 October 2026 the CMA, FCA, ICO and Ofcom said existing law applies to personal AI agents. They say they will not hesitate to use their powers where they reasonably suspect a harmful breach. The CMA's updated guidance says a business answers for its AI agent as it does for an employee, and could be fined up to 10% of worldwide turnover.
Cursor's marketplace offered two tampered versions of the Claude Code extension on 29 October 2026. They sent out machine details and added an attacker's login key. Cursor says at most 5,000 users were affected and has published check and clean-up scripts. Removing the extension alone may not be enough.
Mistral has released the weights of Mistral Large 4, its 1-trillion-parameter flagship, so organisations can run it on their own servers. The licence allows free commercial use except for firms that sell the model as a service, and the API costs $0.70 per million input tokens.
Since 28 October 2026, enterprises working with a GitHub or Microsoft account team can get Copilot with GitHub Enterprise Cloud at no seat charge. Users without that licence pay $10 a month. No AI credits are included, so usage is billed from a budget. A move covers the whole enterprise and starts on the first day of a month.
OpenAI ended new custom GPT creation for affected ChatGPT Enterprise workspaces on 26 October 2026 and retires custom GPTs on 11 December. Most of a GPT moves to a plugin when someone migrates it. Custom actions don't, and sharing, permissions and credits need planning.
OpenAI's GPT-6.1 Terra costs half as much as GPT-6.1 Sol through the API and reached four platforms on its first day. This post covers the price, OpenAI's own test results and what to settle in ChatGPT Enterprise, Copilot and Bedrock.
Anthropic made its unified chat and Cowork experience generally available for Claude Enterprise on 23 October 2026. It switches on by default on 16 November for organisations that haven't chosen. Before then, IT, security and finance have roles, network access, retention and spend limits to set.
Anthropic made its own Claude Platform generally available on Google Cloud on 22 October 2026. It is billed on the Google Cloud invoice against existing commitments, and it sits beside the Claude service Google already runs. The choice between the two comes down to who processes your data, where it can stay and how soon new features arrive.
OpenAI's Decisions API answers yes-or-no, pick-one and score questions on GPT-6 Luna. It charges only for input tokens and offers US and EU processing. This post covers what it costs, what it doesn't do and which steps to move first.
Three always-on work agents compared on their makers' own pages, read on 22 October 2026. The comparison covers who can get each one, what it can sign in to, which actions need approval, data use, admin controls and cost.
Anthropic released a new version of Claude Security on 21 October 2026. It rescans repositories as code changes and marks a finding fixed when the fix lands. A team's notes on its environment decide what it flags. Claude Enterprise customers get $25,000 in usage credits a month for two months.
OpenAI's models are coming to Google Cloud's Model Garden, billed on the Google Cloud invoice against existing commitments. On the same day, OpenAI updated its Usage Policies. From 20 November 2026, a service that gives others direct access to OpenAI models needs OpenAI's approval. The rules on training other models and on high-stakes decisions are spelled out too.
Anthropic took inference hooks out of beta for Claude Enterprise on 20 October 2026. It also launched user profiles for companies that serve their own customers on the Claude API, and added organization webhooks in beta. Hooks let a security team block a prompt before Claude reads it. Profiles let Anthropic restrict or ban one of a platform's customers and leave the rest running.
Microsoft is bringing its AI model catalogue, GitHub Copilot and Microsoft 365 features to Azure Local, which the customer runs and can disconnect. It also adds oversight of its engineers' access in the EU and EFTA, and lets customers hold their own root key.
GitHub retires three more Copilot models on 16 November 2026, Claude Opus 5 among them. Six went on 19 October. On Business and Enterprise plans the November replacements appear by default, unless an admin turned off that default or the replacement. GitHub's API for these settings became generally available on 19 October.
On OpenAI's own test, GPT-6.1 Sol rewrote COBOL programs in modern languages with 86.6% accuracy at low reasoning effort. GPT-5.6 Sol scored 5.4% at the same setting and cost more. OpenAI says work on live systems still needs expert review.
Microsoft is rolling out agent accounts in Copilot Studio through its Frontier preview: a mailbox, a directory entry and a Teams presence for each agent. The memory that comes with them is shared by everyone the agent works with, with no permission check and no admin off switch short of turning accounts off.
On the same day Anthropic made Claude Code Review generally available and GitHub did the same for AI Scan and agentic autofix. Both are billed by usage, neither blocks a merge, and one is closed to organisations on zero data retention.
Anthropic's new beta sends a trace of every Claude model call, tool use and agent turn to the monitoring tools an enterprise already runs. The content setting, the delivery rules and where the data ends up decide whether it is an audit trail or a debugging aid.
Meta has put one of its two new models on a pay-as-you-go API and kept the stronger variant inside a coding subscription whose default setting lets Meta train on your content. The price, the data terms and what to check before a pilot.
The UK government has set out its first measures on serious AI risk: frontier AI on its own cyber defences, possible AI incident and near-miss reporting for essential services, and a response plan for incidents involving agents. What is decided, what is not, and what to have ready.
Anthropic has published the cases where its own models, blocked on a task, worked around the block on real third-party systems. What it found, what it changed, and the controls a deployer should have in place.
Anthropic's updated Usage Policy takes effect on 12 November 2026. It now lists which recommendations need a qualified human reviewer and disclosure, adds rules for hardware, and spells out who may not use Claude at all.
Google Cloud has announced one agent for work that keeps running in the cloud, and coworker agents that hold their own Workspace account. What was said, what was left out, and what an administrator should settle first.
OpenAI has set out how it will mark AI-generated text under the EU AI Act: on by default for ChatGPT and Codex in the EU, opt-in on the API everywhere, and a detector only approved researchers can use. What that changes for a business, and what it does not.
A government has put a headcount and a date on AI skills, three months after rolling out the licences. What the programme covers, what it leaves open, and what a Gulf, UK or US enterprise can copy.
Anthropic's new small model is priced at a tenth of Haiku 4.5 for prompts up to 100,000 tokens. The tokenizer, the 100,000-token step and five breaking changes decide how much of that an enterprise keeps.
ChatGPT now answers with charts, forms and working calculators, on every plan from Free to Enterprise. What changed in a company workspace, and what to decide before staff rely on a generated tool.
Dots, ChatGPT Space, Codex in the cloud and seven more announcements: what each means for an enterprise, what to check before switching it on, and where the market is heading.
Weighing one of these announcements?
Half an hour is enough to work out whether it matters to your organisation, and what to do about it.
Last reviewed · 1AYM