Skip to content
IIAIT

AI Today · 18 November 2026

Microsoft rolls out the new Copilot and its Autopilot agent, and Apple tests daily limits for Siri AI

Microsoft opens Ignite with a rebuilt Copilot and a first autonomous agent for early customers, Apple starts testing daily AI allowances tied to iCloud+, Google documents attackers automating exploits from security patches, OpenAI lets outside safety evaluators inside, and Hugging Face changes how it routes open-model requests.

  1. 01

    Microsoft makes the new Copilot Home and Office in Copilot generally available, and starts rolling out the Autopilot agent

    At its Ignite conference on November 17, 2026, Microsoft made Home generally available as the new starting point in Microsoft 365 Copilot, combining Chat and Cowork, and brought Word, Excel and PowerPoint editing directly into the Copilot app. Autopilot, an always-on agent with its own Microsoft Entra Agent ID, a named manager and its own activity log, began rolling out to customers in the Frontier early-access program, while the app builder Copilot Code is due to become generally available in the coming weeks. The Register notes that the keynote opened with an apology for last week's Microsoft 365 outage.

    What it means for you

    If your company uses Microsoft 365 Copilot, expect the new Home screen and ask IT which features are on in your tenant. Before anyone delegates recurring work to Autopilot, agree who manages each agent, what it may access and which actions it must confirm with a person first.

  2. 02

    Apple tests daily limits for advanced Siri AI and Apple Intelligence features, with higher allowances on paid iCloud+ plans

    The iOS 27.3, iPadOS 27.3 and macOS 27.3 betas released on November 16, 2026 add a free daily allowance per device for features that use Apple's server models, such as multi-step personal requests to Siri, image generation and editing, and some Shortcuts and third-party app features. Basic Siri commands stay unlimited. According to MacRumors and 9to5Mac, the 200GB iCloud+ plan at $2.99 a month gives up to 2x the usage, rising to 20x on the 12TB plan, with no price change; Apple has not published the base limits, which reset around midnight.

    What it means for you

    If your team relies on Siri or Apple Intelligence for drafting or automations, do not build critical workflows on features that can stop for the day. When the update ships, check which iCloud+ plan company devices use, since the 50GB plan adds no extra AI usage.

  3. 03

    Google reports the first AI agent framework it has seen turning security patches into working exploits without human help

    In its quarterly AI threat report published November 16, 2026, Google Threat Intelligence Group and Mandiant describe a framework called exploit_agent, found on an exposed server, that compared web application code before and after security updates, wrote exploits for the fixed flaws and ran them automatically. Google links it to a suspected Russian-speaking operator targeting French media, entertainment and education websites. The report also documents attackers using open-source agent frameworks such as CAIRN and PentAGI in live intrusions, and stealing AI service credentials to resell access to models.

    What it means for you

    The gap between a vendor's security patch and a working attack is shrinking, so apply updates to internet-facing software quickly rather than waiting for the next maintenance window. Treat API keys for AI services like any other credential: store them securely, rotate them and watch for unusual usage on your bill.

  4. 04

    OpenAI lets four outside safety groups work inside the company, starting with a four-month Apollo Research pilot

    OpenAI said on November 16, 2026 that evaluators from Apollo Research, Goodfire, METR and Transluce will work inside the company in separate pilots. Apollo will place six researchers at OpenAI for four months to check whether models try to undermine their own safety controls during training, testing and internal use, and plans two public reports. According to TNW, OpenAI may ask to redact confidential evidence but cannot remove Apollo's conclusions, while Apollo calls the access a first step short of what it wants.

    What it means for you

    Nothing changes in the ChatGPT or API products you use today. If your company evaluates AI vendors, independent testing that starts before release is a useful question to add to your due diligence, alongside how the vendor publishes the results.

  5. 05

    Hugging Face switches language model requests on Inference Providers to OpenRouter's routing, ending custom provider keys for them

    Hugging Face announced on November 16, 2026 that requests to chat models on its Inference Providers service now go through OpenRouter, which picks a provider for each call based on live price, speed and reliability and retries on another provider if one fails. Existing tokens and code keep working, prices stay at the provider's rate with no Hugging Face markup, and users can still force a provider or exclude one in settings. The breaking change is that customers can no longer use their own provider API keys for these models; image, video and audio generation are unaffected.

    What it means for you

    If your developers call open-weight models through Hugging Face with their own provider keys, those integrations need attention now and billing moves to your Hugging Face account. Review the provider list in your organization settings and turn off any provider your data policy does not allow.