Max Laktsionau, Forward Deployed Engineer at AdvantageWorks Max Laktsionau 7 min read AI-assisted

Claude Fable 5.2, ChatGPT Wallet and the rest of AI's week (Nov 9 – Nov 16)

Printed AdvantageWorks AI Weekly for Nov 9 to Nov 16, 2026 on a desk, with a sticky note about Claude Fable 5.2

Agents got a sanctioned way to pay this week and lost a way to book. Anthropic and Google both shipped faster, cheaper tiers, and a day-long Microsoft outage showed how much of the working day, Copilot included, hangs on one email service.

Anthropic ships Claude Fable 5.2 and Mythos 5.2 at a lower price

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, the same underlying model with different safeguards. Fable 5.2 is open to all users and costs $8 per million input tokens and $40 per million output tokens, with an optional fast mode at double that. Mythos 5.2 stays behind Anthropic's cyber and life-sciences verification programs. Anthropic says Fable 5.2 at High effort matches Opus 5.5's best FrontierCode score for roughly half the cost per task, and AWS put both models on Amazon Bedrock the same day .

The detail that bites in production is buried in the AWS post. Fable 5.2's classifiers block most security and biology tasks, and on Bedrock the retry on Opus 4.8 or Opus 5 is your code's job, not an automatic switch. A pipeline that runs dependency audits or vulnerability triage through Fable will start returning refusals the day someone flips the model ID, so wire the fallback (and the cache credit token AWS returns with the refusal) before the switch. Anthropic itself still recommends Opus 5.5 as the daily default, which makes Fable a per-task upgrade you route hard work to while Opus stays the house model.

Google puts Gemini 4 Hydrogen in Search, Antigravity and GitHub Copilot

Google introduced Gemini 4 Hydrogen on November 12 as its fast execution model, priced flat at $1.50 per million input tokens and $7.50 per million output tokens across a 1 million token context. It is rolling out across the Gemini API, Gemini for business, the Gemini app and Google Search, and Google added an Orchestrator to Antigravity that switches between Carbon and Hydrogen within one session. A day later GitHub made Hydrogen generally available in Copilot for Pro, Business and Enterprise plans.

Over-the-shoulder view of VS Code with the GitHub Copilot model picker open on Gemini 4 Hydrogen

Check Copilot's default this week. New models are enabled automatically unless an administrator has turned off the global default, and Hydrogen is billed at provider list pricing under usage-based billing. So it may already sit in your developers' model picker, spending against your budget, before anyone has run it on your codebase. Open the model policy in Copilot settings , decide whether new models wait for a short internal trial, and write that rule down so the next release does not reopen the question.

OpenAI launches ChatGPT Wallet so its agents can pay at checkout

OpenAI introduced ChatGPT Wallet on November 13, built with American Express, Mastercard and Visa. The user approves each purchase with a check such as Face ID, and the network issues a payment token so neither the agent nor the merchant sees the real card. It starts in OpenAI's dots agents for ChatGPT Pro subscribers in the US, and TechCrunch notes that OpenAI plans spending limits and category rules later.

Packing slip showing a ChatGPT Wallet agent payment on a box being taped at an online shop's packing table

Companies that sell online have homework in OpenAI's own launch post. Its Commerce Partner Playbook asks merchants to let signed agent requests (Web Bot Auth) through their CDN or firewall, and it warns that fraud systems may decline legitimate agent purchases. Ask your payment provider this month how it scores a tokenized agent checkout. OpenAI names Stripe, Adyen and Checkout.com as partners, and a decline your risk rules generate is revenue you never see in a report.

United Airlines blocks AI agents from booking flights

United said on November 11 it will restrict unauthorized third-party AI agents from booking and managing reservations, both on its own platforms and through travel agencies and aggregators. The airline cited the risk of customers sharing account credentials with assistants and federal rules that require a human to complete check-in. CBS News reports that Southwest went the other way and made its inventory viewable to ChatGPT users.

Traveller reading the hazardous materials placard at a United Airlines check-in counter in an airport

United's statement describes a pattern that reaches well past airlines. Any internal agent that logs into a supplier portal, a travel tool or a bank with an employee's password is doing exactly what United just banned, and the site owner can switch it off overnight. List those workflows now , note which ones have an official API or partner channel behind them, and treat the rest as on loan.

A day-long Microsoft outage takes down Outlook, Teams and Copilot

Microsoft said on November 14 that services had fully recovered after a global outage that hit Teams, Outlook and Copilot for more than 24 hours. Reuters reports the fault sat in features tied to Exchange Online, Downdetector logged more than 7.4 million reports, and Microsoft ruled out malicious activity and AI as the cause. The Guardian adds that the company traced it to an earlier network capacity upgrade.

Office whiteboard saying Outlook and Teams are down with phone extensions, above a desk phone off its cradle

Inbox triage, meeting summaries that feed the CRM and Copilot drafting all stopped together, because they all start in Exchange. Organizations running Microsoft services on their own servers were somewhat spared, Reuters says, which most mid-market firms cannot copy. What you can copy is a written fallback : which AI-assisted processes queue work and replay it after recovery, which switch to a manual path, and who decides. Morningstar DBRS put insured losses at $300 million to $500 million, so check what your own cyber and business-interruption cover says about a vendor outage.

OpenAI arrives on AWS through Marketplace and Bedrock

On November 9 OpenAI made ChatGPT Enterprise available in AWS Marketplace , so companies can buy it on their AWS bill and apply eligible spend toward AWS commitments, with the same features as a direct contract. AWS followed on November 11 with a guide to running OpenAI Codex on Amazon Bedrock , where Codex stays on the developer's laptop but its model calls go through Bedrock under IAM Identity Center.

Highlighter marking a ChatGPT Enterprise AWS Marketplace line on a printed AWS invoice in a meeting

Unspent AWS commitments now have a new outlet, and that changes the vendor-consolidation conversation with finance . The Codex guide is the more useful read for the platform team, because it lays out three places to enforce policy: native AWS access for a small pilot, AgentCore Gateway for managed routing and guardrails, or your own LLM gateway in the request path. Pick one before the pilot grows, because moving model access between those paths later means re-issuing every developer's credentials.

Also this week

  • Jensen Huang resigns as Nvidia CEO. Huang stepped down on November 12 after reports of extramarital affairs, and CFO Colette Kress became interim CEO. Nvidia shares fell as much as 14.3% before closing down 4.3%, Reuters reports. Reuters
  • Anthropic sets its IPO price range. Anthropic plans to sell 825 million shares at $70 to $78 on the Nasdaq under the ticker ANT, with pricing expected on November 18 and trading on November 19, CNBC reports. CNBC
  • Google launches the Gemini Decisions API. A new endpoint scores up to 1,000 yes/no, choice and rubric questions in one pass at $0.10 per million input tokens, with no output tokens billed. Google
  • IBM and Microsoft announce Imagine Enterprise. The agent platform pairs IBM Consulting's more than 6,000 prebuilt agents and assets with Microsoft's cloud and security stack, aimed at regulated industries. Microsoft
  • The US and its allies agree to protect AI supply chains from China. Pax Silica partners including Japan, the UK and India will build a joint list of AI technologies and supply chains to protect. Techmeme, citing the Financial Times
  • The FAA rules out its SMART AI system as the cause of a nationwide ground stop. Officials blamed overloaded switches and routers in aging telecom equipment for the four-hour stop on November 13. AVweb
  • Google Cloud launches Gemini for healthcare. The preview connects records, billing and insurance systems for tasks such as prior authorization review, with HCA Healthcare, Highmark Health and Northwell Health as launch partners. Google
  • Apple is preparing Harbor, a Mac app for running local models. A leak shows it pooling memory across Thunderbolt 5 Macs to serve open-weight models on-device, currently in private preview. 9to5Mac

Frequently asked questions

As of November 2026, Claude Fable 5.2 costs $8 per million input tokens and $40 per million output tokens through Anthropic's API, which is lower than Claude Fable 5.1.

An optional fast mode, which Anthropic says speeds up generation by up to 2.5 times, costs $16 per million input tokens and $80 per million output tokens. Claude Mythos 5.2 is the same underlying model with fewer safeguards and is available only through Anthropic's cyber and life-sciences verification programs.

The list price is only half of the cost question. Anthropic reports that Fable 5.2 at High effort matches Opus 5.5's best score on its FrontierCode coding benchmark for roughly half the cost per task, while it still recommends Opus 5.5 as the default for most coding and knowledge work. Compare cost per finished task on your own workload, not price per token.

No, not in its standard form. Claude Fable 5.2 ships with classifiers that block most cybersecurity tasks (penetration testing, exploit generation, binary vulnerability work) and most research-biology tasks.

Anthropic routes that work to Claude Mythos 5.2, which organisations can apply for through its verification programs. On Amazon Bedrock, AWS says a declined request can be retried on Opus 4.8 for security queries or Opus 5 for biology queries, but the retry is your code's job rather than an automatic switch. If the declined request hit the prompt cache, the refusal includes a fallback credit token so the retry is billed as cache reads.

Teams that run dependency audits or vulnerability triage through Claude should build and test that fallback path before changing the model ID to Fable 5.2.

ChatGPT Wallet is OpenAI's payment feature, launched on November 13, 2026, that lets ChatGPT's dots agents pay for purchases on a user's behalf without seeing the user's real card details.

It was built with American Express, Mastercard and Visa. The user adds a card once, reviews the items, total and card before each purchase, and approves it with a security check such as Face ID or a fingerprint. The card network then issues a payment token in place of the real card number, and the user keeps the card's rewards and dispute protections.

As of November 2026, it is available only to ChatGPT Pro subscribers aged 18 and over in the US with eligible US-issued cards. OpenAI says it plans to add spending limits, category rules and time-based pre-approvals later.

Merchants mainly need to make sure their site and fraud systems do not block legitimate agent purchases. ChatGPT Wallet's payment tokens work with existing checkout flows.

OpenAI's Commerce Partner Playbook lists the steps:

  • configure the CDN or firewall to recognise signed agent requests using Web Bot Auth;
  • ask payment and risk providers how they recognise agent purchases, since fraud rules may decline them;
  • claim the business in ChatGPT and upload a product feed with current pricing and availability;
  • optionally build a plugin or add WebMCP tools so agents can check availability and manage carts.

OpenAI names Stripe, Adyen and Checkout.com among the payment providers, and Forter, Riskified, Sardine and Signifyd among the fraud-prevention partners, that it is working with.

Gemini 4 Hydrogen is available across Google's free and paid tiers, but API use is paid: $1.50 per million input tokens and $7.50 per million output tokens, flat across its 1 million token context window, as of November 2026.

Google says Hydrogen is coming to all tiers of Google AI Studio, the Gemini API, Gemini for business, the Gemini app and Google Search, and to paid Google AI Pro and Ultra subscribers in Google Antigravity.

In GitHub Copilot, Hydrogen is generally available on Pro, Pro+, Max, Business and Enterprise plans and is billed at provider list pricing under usage-based billing. Copilot enables new models by default unless an administrator turns off the global default, so Business and Enterprise admins should check the model policy in Copilot settings.

United Airlines said on November 11, 2026 that it will restrict unauthorised third-party AI agents from booking flights and managing reservations, citing cybersecurity risk and federal aviation rules.

The airline warned that sharing account credentials with personal AI assistants can expose sensitive information and lead to unintended purchases, account changes or use of MileagePlus miles. It also said agents completing steps such as check-in could breach rules that require a traveller to confirm their bags hold no banned items. The restriction applies on United's own platforms and through travel agencies and aggregators.

Other airlines took a different line. CBS News reports that Southwest made its flight inventory viewable to ChatGPT users. For companies, the lesson is wider than travel: any agent that logs into a third-party site with an employee's password can be switched off by that site.

No. Microsoft said the outage that took down Outlook, Teams and Copilot for more than 24 hours from November 12, 2026 was not related to cyberattacks or AI agents.

According to The Guardian, Microsoft traced the incident to an issue introduced during an earlier network capacity upgrade. Reuters reports that it hit features tied to Exchange Online, that Downdetector logged more than 7.4 million reports worldwide, and that services were fully restored on November 14.

Copilot went down with Exchange because it depends on the same services. Organisations that run AI-assisted email and meeting workflows on Microsoft 365 should document which processes queue and replay work after an outage and which switch to a manual path.

Yes. Since November 9, 2026, ChatGPT Enterprise has been available in AWS Marketplace, so organisations can buy it through their AWS billing relationship and apply eligible spend toward their AWS commitments.

OpenAI says AWS Marketplace customers get the same models, features and functionality as buying directly, with new features released at the same time. ChatGPT for Healthcare and ChatGPT for Financial Services are also available there to eligible customers.

Separately, developers can run OpenAI models, including through Codex, on Amazon Bedrock. AWS's guide describes three control points for that: native AWS access with IAM Identity Center for small pilots, AgentCore Gateway for managed routing and guardrails, or an LLM gateway the platform team runs in the request path.

Editorial statement

This article was produced with AI assistance and has undergone human review and editorial control. Max Laktsionau holds editorial responsibility for its content within the meaning of Article 50(4) of Regulation (EU) 2024/1689.