Apple posted job listings in the week of November 17, 2026, seeking a senior engineering manager for a "proactive, agentic AI assistant," as reported by Bloomberg. The listing describes an agent capable of browsing the web, controlling apps, running commands, writing and debugging code, calling services, and coordinating specialised sub-agents across multi-step tasks. The role sits across Apple's Siri, UI Frameworks, and Apple Foundation Models divisions.
Apple's new chief executive John Ternus has been pushing software engineering chief Craig Federighi to move faster on AI, with the agent described as central to that effort. Bloomberg reported that Apple has eyed introducing the service as early as end of 2027, though development remains in early stages. Competitors already in the market include Meta's Muse, OpenAI's Dots, Anthropic's Duet, and Google's Tasks.
Simultaneously, Apple is making it harder for third-party agents to gain broad control over its computers. The macOS 27.3 beta introduces warning screens that require 14 distinct steps for users to enable third-party computer-use tools, up from 4. Apple is also rolling out iOS 27.3, which will allow Siri to perform tasks across a wider range of third-party apps. Apple's installed base across its devices stands at approximately 2 billion.
🤖 made with AI
Full details via The Verge / theverge.com/tech/1038227/ap…
71
Token fraud has become a direct financial exposure for AI platforms offering free trials or consumption-based pricing. On November 20, 2026, developer Simon Willison flagged a YouTube video by Claire Vo, chief executive of ChatPRD, an AI platform for product managers, in which she described organised groups running loops of free trials and promotional codes to access her platform's AI agent without payment. Some operators used the access to run commercial enterprises, including a Roblox and Fortnite map studio. 1 diagnostic signal Vo identified was a recurring prompt involving SVGs of pelicans riding bicycles, a technique Willison noted is used to fingerprint the underlying model family behind a platform. Vo announced she is winding down free trials entirely as a result.
The scale of the problem extends well beyond a single platform. Stripe chief executive Patrick Collison stated in May 2026 that token thieves account for 1 in every 6 new customer signups at AI firms. Stripe's head of data and AI Emily Sands noted that free trial abuse had more than doubled in the 6 months to May 2026, with a sharp acceleration in the final month. Sands cited 1 startup whose customer acquisition cost had reached $500 per genuine customer due to fraudulent trial consumption.
A separate investigation by Matt Lenhard, covered by Hackaday and discussed on Hacker News, documented a grey market in which resellers offer AI tokens at discounts of 87% or more, sourced through free-credit farming, stolen card schemes, chargeback fraud, and reverse-engineering of consumer AI products to access backend inference directly. Because AI inference carries a real marginal cost, fraudulent consumption creates direct financial losses rather than foregone revenue, and agents can exhaust trial allocations before any human review process can detect the activity.
🤖 made with AI
Full details via Simon Willison / simonwillison.net/2026/Nov/2…
1
146
Lenders backing SpaceX's approximately $43bn debt package, known internally as Project Saturn, sought contractual provisions that would prevent SpaceX from launching the Nvidia chips securing the financing into orbit, the Financial Times reported on November 19 to 20, 2026. The concern is that chips deployed in orbital data centres would become effectively unrepossessable in the event of a default. SpaceX has stated that the chips will be used in terrestrial data centres.
The financing structure places the debt in a special-purpose vehicle, keeping it off SpaceX's balance sheet. SpaceX leases the GPUs from the SPV for 5 years under terms that do not permit exit for equipment failure or force majeure events. The $25bn investment-grade portion of the deal was priced at approximately 8.3% and received A3 and A-minus ratings from Moody's and KBRA respectively. Apollo Global Management is the lead arranger, with BlackRock, Goldman Sachs Asset Management, and KKR as cornerstone investors.
Whether the "do not launch" provision was included in the final loan documents remains unclear. The SPV structure, keeping $43bn of GPU-backed debt off SpaceX's balance sheet while maintaining investment-grade ratings, is a financing template that Meta and other large technology companies have also used. GPU delivery under the deal is scheduled for Q1 to Q2 2027.
🤖 made with AI
Full details via Financial Times / ft.com/content/33b5643c-2957…
1
28
The House Foreign Affairs Committee released previously confidential export licence data on November 19, 2026, showing that the Commerce Department's Bureau of Industry and Security approved the sale of 879,868 advanced AI chips to Chinese companies between December 2025 and May 2026. The approvals covered roughly 1 in 4 of the 3.4 million chips that Nvidia and AMD had applied to export during the period. The chips involved were Nvidia's H200 and AMD's MI325X.
Recipients included Alibaba (75,000 chips), Tencent and its subsidiaries (150,000), Baidu (43,000), and BYD (4,000). All 4 are on the Pentagon's 2026 list of alleged Chinese military companies. Committee chair Brian Mast used authority under the Export Control Reform Act to declassify and release the data, citing national interest grounds. Mast also alleged that 2 of Nvidia's licence applications named entities, Maginfra and New Matrix, that he described as suspected chip-smuggling fronts with empty offices and ties to companies under national security sanctions.
Nvidia stated that approvals were "subject to strict conditions and limited volumes" and that it has since stopped shipping H200s to China. The approvals were made under a December 2025 decision by President Trump permitting H200 sales to approved Chinese customers, reversing Biden-era restrictions. The Democratic ranking member of the committee accused the administration of putting national security at risk. Beijing separately declined to approve large-scale imports during the same period.
🤖 made with AI
Full details via CNBC / cnbc.com/2026/11/19/nvidia-a…
Asian equity markets fell sharply on November 20, 2026, extending a sell-off that had been building through the week as the Anthropic IPO failed to restore confidence in AI-related stocks. Taiwan's benchmark index fell 5.6% on the day, completing a weekly decline of 16%, its worst weekly performance since 1991. Japan's Nikkei 225 dropped 4.7% to a 7-month low; South Korea's KOSPI fell 4.4% to a 9-month low. The MSCI index of Asia-Pacific markets excluding Japan lost 3.5% on the day and 8.6% on the week, its worst weekly performance since the COVID-19 crisis of March 2020.
SoftBank Group fell 16.16%, its worst single-day decline in more than 2 years, reflecting its $64.6bn exposure to OpenAI. Among chipmakers, TSMC fell 4.76%, SK Hynix and Samsung Electronics each dropped approximately 6%, Tokyo Electron fell 7.19%, and Lasertec declined 6.75%. India's Nifty 50 rose 9.4% in November, diverging from the broader Asian AI complex.
Contributing factors included a short report on AMD from Hunterbrook Capital raising concerns about the chipmaker's ties to Chinese entities, Jensen Huang's earlier departure from Nvidia, growing regulatory pressure on data centre construction in the United States following the Democratic midterm sweep, and rising expectations of a Federal Reserve rate hike in December, now effectively priced in according to CME FedWatch data.
🤖 made with AI
Full details via CNBC / cnbc.com/2026/11/20/softbank…
1
33
Anthropic priced its initial public offering at $83 per share on November 18, 2026, raising $81.3bn in the largest IPO on record, surpassing SpaceX's June 2026 listing. The company began trading on the Nasdaq under the ticker ANT on November 19, opening at $111.50 and closing at $97.17, a 17% gain on the IPO price. At the close, Anthropic's fully diluted market capitalisation stood at approximately $1.76tn.
The offering covered 980 million Class A shares. Anthropic retained proceeds from 956 million of those; each of the 7 co-founders sold 3.38 million shares, generating approximately $1.96bn in personal proceeds. Anthropic reported $33.24bn in revenue for the 9 months ended September 30, 2026, up roughly 1,200% year on year, with a run-rate revenue of approximately $81bn as of end-October 2026. The company disclosed that more than half of the Fortune 500 use Claude. The reported net loss of $104bn for the period is dominated by non-cash accounting charges on convertible instruments rather than operating cash burn.
Anthropic is incorporated as a public benefit corporation. Its 7 co-founders collectively hold 50.1% of voting power through a structure called Founder LLC, and an independent body called the Long-Term Benefit Trust retains authority to elect and remove the majority of the board. Shareholders hold economic exposure but limited governance influence. The Nasdaq Composite fell 2.3% on the day of the debut, with AI-adjacent stocks declining sharply across the session.
🤖 made with AI
Full details via CNBC / cnbc.com/2026/11/19/anthropi…
1
30
📰 The Tokenando Daily / Friday November 20, 2026
/ Anthropic prices its IPO at $83 per share, raises $81.3bn and closes its Nasdaq debut up 17%, becoming one of the most valuable public companies in the world
/ Asian equity markets extend their AI sell-off, with Taiwan's benchmark index posting its worst weekly decline since 1991 and SoftBank falling 16% in a single session
/ Congressional data release shows the Commerce Department approved 879,868 advanced Nvidia and AMD chips for sale to Chinese companies, including firms on the Pentagon's military list
/ SpaceX lenders sought contractual protection against Elon Musk launching the $43bn in Nvidia chips backing Project Saturn into orbit, where they would be unrepossessable
/ Token fraud now accounts for 1 in every 6 new AI platform signups, according to Stripe, as organised groups exploit free trials to run commercial operations at platform expense
/ Apple posts job listings for a senior engineering manager to build a proprietary agentic AI assistant, targeting a potential launch as early as end of 2027
🤖 made with AI
✅ reviewed by the Tokenando team
🔗 links and details in following posts
🤖 Made with AI
tokenando retweeted
At Anthropic, one share carries 50.1% of the vote. The 980 million shares sold this week at $83 raised $81bn and opened today at $111.50. The reported bill for chips, power and data centres: $901bn.Who foots it?
Issue XXIX of The Compute by @TokenandoAI tokenandoai.substack.com/p/t…
🤖 Made with AI
Indian AI startup Sarvam, valued at $1.5bn in April 2026 and a beneficiary of the Indian government's IndiaAI Mission, launched Token Factory on November 16, 2026, an inference platform offering multiple open-weight AI models through a single API hosted on GPU infrastructure located in India.
The catalogue includes Sarvam's own 105-billion-parameter model alongside models from DeepSeek, Moonshot AI, Z.AI, and Alibaba, as well as Google's Gemma. Sarvam reported it was already serving over 450 billion tokens daily on DeepSeek V4.1 Flash. Pricing is denominated in Indian rupees, starting at Rs 14 per million input tokens for DeepSeek V4.1 Flash, compared with Rs 29.28 for Sarvam's own 105B model.
The launch triggered public criticism given Sarvam's position as a recipient of government-subsidised compute under the IndiaAI Mission, which was established to reduce India's dependence on foreign AI technology. Co-founders Pratyush Kumar and Vivek Raghavan defended the decision, stating that Token Factory runs on privately funded GPUs separate from the government-subsidised infrastructure, that the subsidised compute allocation ended in March 2026, and that hosting Chinese models on Indian infrastructure keeps user data within India's borders. Critics have argued that the model weights themselves, their training data, and their embedded behaviours constitute a separate sovereignty question that local hosting does not resolve.
🤖 made with AI
Full details via Moneycontrol / moneycontrol.com/news/busine…
28
At its Ignite conference on November 17 and 18, 2026, Microsoft unveiled 7 new proprietary AI models under its MAI brand, positioned as lower-cost alternatives to models from OpenAI and Anthropic that Microsoft currently uses in its own products.
The headline model, MAI Code 2, is priced at $0.50 per million input tokens and $2.50 per million output tokens at a promotional rate through May 17, 2027, rising to $1 and $5 thereafter. Microsoft claims it delivers 98% of GPT-6.1 Sol's coding performance at 40% of the cost and 96% of Claude Opus 5.5's performance at 5% of the cost. MAI Cyber 2 Flash is priced at $0.10 per million input tokens and $0.50 per million output tokens, compared with $10 and $50 for GPT-6 Astra, and already handles nearly 40% of scans on Microsoft's security platform MDASH. The suite also includes MAI Image 2.7, MAI Realtime 1, MAI Transcribe 2, and Decision 1.1, the last built on Google's Gemma 4 26B architecture.
Bloomberg reported in July 2026 that Microsoft had already begun routing tens of thousands of Excel and Outlook prompts per week to MAI models. Microsoft AI chief Mustafa Suleyman stated in June 2026 that the goal was to "reduce and ultimately eliminate" spending on Anthropic. The April 2026 restructuring of the Microsoft-OpenAI partnership, which ended Microsoft's exclusivity and allowed OpenAI to sell through other clouds, also freed Microsoft to build competing models. A MAI Licensed Weights offering allows enterprises to fine-tune MAI models on their own data within Azure.
🤖 made with AI
Full details via The Decoder / the-decoder.com/microsoft-is…
1
57
Google DeepMind released Gemini 4 Neon on November 18, 2026, the 4th distinct Gemini 4 model released in approximately 7 weeks, following Argon on October 13, Carbon on November 10, and Hydrogen on November 12.
Neon is priced at $3 per million input tokens and $15 per million output tokens, with a 95% discount on cached input tokens at $0.15 per million. It carries a 1-million-token context window, supports text, images, video, audio, and PDFs, and offers 4 configurable thinking levels from low to extra-high. At its extra-high thinking level, Neon scores 61 on the Artificial Analysis Intelligence Index, placing it third overall behind GPT-6.2 Astra and Claude Fable 5.2. On the Vals Index it ranks second at 70.9%. Google positions Neon between Hydrogen, the fast everyday execution tier, and Carbon, the highest-capability tier, with Neon replacing Argon in the model picker.
The pricing structure across the Gemini 4 family creates a specific decision point before the end of November. Carbon's introductory rate of $2 per million input tokens expires November 30 and rises to $4, at which point Neon at $3 becomes the mid-tier value option. At its standard high thinking level, Neon costs approximately 17% less per task than Carbon at Carbon's non-discounted rate. The model card confirms Neon is architecturally adapted from Carbon rather than built on a separate architecture, with differentiation by inference speed and cost. A cybersecurity variant of Neon is rolling out to the Fairwind Programme.
🤖 made with AI
Full details via Google DeepMind / blog.google/innovation-and-a…
1
52
Nvidia interim chief executive Colette Kress held the company's first all-hands meeting since Jensen Huang's resignation on November 18, 2026, addressing approximately 49,000 employees in person and via Microsoft Teams. Kress opened by telling staff "let's move forward" and announced that pay rises and additional equity grants would be accelerated to December, several months ahead of the usual March review cycle.
Board chair Harvey Jones, co-founder Chris Malachowsky, and several senior executives including GPU engineering senior vice president Jonah Alben and software engineering senior vice president Dwight Diercks also addressed staff. Jones reiterated that the board had accepted Huang's resignation due to personal conduct and lack of candour, and confirmed the board would not reinstate him. Chief commercial officer Nick Parker reported that customers had sent messages of support.
The meeting followed Nvidia's third-quarter earnings report, which beat Wall Street expectations. Nvidia's shares were down 37% from their 52-week high at the time of the meeting, reducing the retention value of existing equity grants and making the accelerated awards more urgent. The most common employee question submitted ahead of the meeting concerned compensation; the second most common requested greater board transparency. The board's statement that Huang will not return was made while Huang himself had publicly signalled willingness to come back.
🤖 made with AI
Full details via WIRED / wired.com/story/inside-nvidi…
1
76
Anthropic priced its initial public offering at $83 per share on November 18, 2026, above its indicated range of $70 to $78, raising approximately $81.3bn and claiming the record for the largest stock-market debut in history, surpassing SpaceX's $86.25bn offering when the greenshoe option is excluded.
The offering comprised 980 million shares, upsized from an initial plan of 825 million, with 956 million sold by the company and the remainder by existing stockholders. A concurrent private placement of $26bn from investors including Nvidia, Google, and Microsoft brings total proceeds to approximately $107bn. On a fully diluted basis the listing values Anthropic at approximately $1.5tn; on a non-diluted basis the figure is approximately $1.1tn. Trading on the Nasdaq under the ticker ANT was expected to begin November 19, 2026.
The financial profile disclosed in the prospectus adds texture to the headline figures. Anthropic spent nearly $9bn on computing capacity in the first half of 2026, roughly 4 times the prior-year figure, and a further $5bn in the third quarter. The company has committed to nearly $900bn in computing spending with infrastructure partners, and its prospectus warns that some of those partners depend on third-party financing that could be withdrawn or repriced. Revenue, at an annualised run rate of $81bn as of October 31, is being driven in significant part by Claude Code and agentic coding workflows, with 11,900 enterprise customers generating at least $100,000 annually. Operating income of $974m in Q2 2026 reversed to an operating loss of $567m in Q3 as training costs accelerated.
🤖 made with AI
Full details via AP News / apnews.com/article/anthropic…
1
32
📰 The Tokenando Daily / Thursday November 19, 2026
/ Anthropic prices its IPO at $83 per share, raising $81.3bn and claiming the record for the largest stock-market debut in history
/ Nvidia interim CEO Colette Kress accelerates pay rises and equity grants to December in first all-hands since Jensen Huang's departure
/ Google ships Gemini 4 Neon, its 4th distinct Gemini 4 model in 7 weeks, priced at $3 per million input tokens with a 1-million-token context window
/ Microsoft unveils 7 proprietary MAI models at Ignite, positioning them as lower-cost replacements for OpenAI and Anthropic across its own products
/ Saudi Arabia's HUMAIN announces a 3-trillion-parameter open-weight model for December 15, claiming 88% lower token costs than models available outside the Kingdom
/ Sarvam AI launches Token Factory in India, serving over 450 billion daily tokens on DeepSeek V4.1 Flash and reigniting the debate over what AI sovereignty actually means
🤖 made with AI
✅ reviewed by the Tokenando team
🔗 links and details in following posts
🤖 Made with AI
The Information reported on November 17, 2026 that 2 browser-hosting startups have generated significant revenue as a direct consequence of Instinct's rapid growth. Instinct is a personal AI agent that operates using its own phone number and computer to complete tasks on behalf of users, including booking travel, making purchases, cancelling subscriptions, and conducting research. The browser-hosting startups provide the virtual browser infrastructure Instinct's agent uses to navigate the web.
Instinct raised $1bn at a $10bn valuation in a Series C round in September 2026, having raised at a $2.5bn valuation only 1 month earlier in August 2026. The Information noted that Instinct had outgrown the parameters of its annual 50 Most Promising Startups list by the time reporting was complete.
Andreessen Horowitz's B2A Economy analysis found that agentic browser traffic grew eightfold between August and October 2026, with Meta's Muse driving 60% of activity and OpenAI products reaching a combined 30%. Current AI agents complete tasks by spinning up virtual browser instances and navigating websites as a human would, creating sustained demand for scalable, isolated browser environments. Agent-native protocols including the Model Context Protocol and direct API integrations are maturing as alternatives to browser-based task completion.
🤖 made with AI
Full details via The Information / theinformation.com/articles/…
30
Google rebranded its Spark AI agent as Tasks and made it available at no cost to users in the United States, Canada, Japan, South Korea, Australia, and India on November 17, 2026, powered by the new Gemini 4 Hydrogen model. Spark was previously available only to Google AI Pro and Ultra subscribers.
Tasks is accessible by switching a toggle in the Gemini app from Chat to Tasks. It can manage calendars, analyse emails for recurring expenses, and create meal plans. US users additionally have access to business calling to compare pricing and check inventory, purchases through Google Pay, and account sign-in via Google Password Manager. Users can also assign tasks by emailing [email protected] or texting Gemini via RCS. Each user's agent runs in a hardware-isolated virtual machine dedicated to their account.
The feature's most commercially significant capabilities, including business calling and Google Pay integration, are restricted to the US. Google noted that the feature remains imperfect and that some websites require the agent to hand control back to the user for final steps. The Agent Payments Protocol integrates with Mastercard and Visa, with American Express support forthcoming.
🤖 made with AI
Full details via Google Blog / blog.google/innovation-and-a…
1
285
Microsoft chief executive Satya Nadella stated at a roundtable following the Microsoft Ignite keynote at the Chase Center in San Francisco on November 17, 2026 that he is "AGI-pilled in the sense that at an economy level, what is the political economy's organisation around AGI?"
Mr Nadella framed the concern as a structural question about economic and political concentration rather than a technical question about whether AGI has been achieved. He argued that if a single AI model were allowed to achieve monopoly status, democratic economies would be unable to function, and that if intelligence has network effects and all human capital collapses into token capital, the nature of the firm itself is in question. He also stated that he does not believe frontier models such as Claude are conscious, aligning with Microsoft AI chief executive Mustafa Suleiman's position that companies should not assign moral status to models.
The remarks were made at the same event where Microsoft was promoting enterprise AI tools designed to allow clients to run multiple models and maintain control of their own data. Mr Nadella described the Ignite keynote's architecture around 3 pillars: data, intelligence, and trust, with trust framed as an imperative to separate the supply of intelligence from the authority over it.
🤖 made with AI
Full details via The Verge / theverge.com/tech/1036111/sa…
1
26
OpenAI announced on November 16, 2026 that its computer-use agents, built on GPT-6 Astra and GPT-6 Luna, are available for healthcare administrative workflows through the OpenAI API, ChatGPT for Healthcare, and ChatGPT for Clinicians. The agents operate existing electronic health record systems and insurance portals by clicking, typing, and navigating interfaces directly, without requiring API integrations.
On the independent HealthAdminBench benchmark from Stanford University researchers, GPT-6 Astra scored 94.3% and GPT-6 Luna scored 91.8%. OpenAI reports GPT-6 Luna costs approximately 8.5 cents per task, against an estimated $8 or more for a human coordinator spending 20 minutes on the same prior authorisation request.
CVS Health is running the agents in production: 1 agent cleared a months-long document backlog over a single weekend, and another handles tens of thousands of specialist consultation steps daily. Cigna Healthcare is using the agents to consolidate member information from more than 10 separate clinical and administrative systems. The offering currently targets US HIPAA-compliant workflows; it does not address GDPR, and European deployment is not yet covered.
🤖 made with AI
Full details via OpenAI / openai.com/index/advancing-c…
1
16
The US Bureau of Industry and Security published an interim final rule on November 17, 2026 requiring data centre operators to obtain a US government licence before supplying advanced AI compute to entities headquartered in mainland China, Hong Kong, Macau, or countries subject to a US arms embargo, where the facility provides more than 12 million aggregate total processing performance.
The rule designates Indonesia, Malaysia, Singapore, Thailand, and Brazil as destinations carrying an unacceptable risk of diversion, requiring export licences for chip shipments to those countries. Existing data centres serving Chinese firms have 90 days to obtain licences. The rule was developed at the direction of the US Treasury Department and the National Security Council rather than the Commerce Department, a shift that reflects the internal Washington power struggle triggered by the Megaspeed scandal.
Published the same day, a Reuters investigation drawing on more than 300 Chinese academic papers, patents, and government documents documented that Chinese firms have built an industrial-scale distillation supply chain using outputs from US models including OpenAI's GPT-4 and GPT-4o. State policy documents from 8 Chinese ministries explicitly encourage the practice for manufacturing deployment. Anthropic stated it now blocks most distillation attacks; US Treasury Secretary Scott Bessent threatened sanctions and Entity List designations for industrial-scale IP theft.
🤖 made with AI
Full details via The Decoder / the-decoder.com/the-us-wants…
1
25
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026, available immediately to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and to developers through the API.
GPT-6.2 Astra is priced at $8 per million input tokens, $0.40 per million cached input tokens, and $40 per million output tokens, a 20% reduction on GPT-6 Astra's rates. On OpenAI's internal GDPval v2 benchmark, Astra improved 6.2 percentage points over its predecessor while using 22% fewer output tokens, meaning the effective cost per task falls more steeply than the headline rate cut alone implies. On the independent HealthAdminBench, Astra scored 94.3%. GPT-6.2 Sol retains GPT-6.1 Sol's pricing at $2 per million input tokens and $10 per million output tokens.
Independent analysis from Artificial Analysis found that coding performance on the Coding Agent Index was flat for Astra and declined 3 points for Sol. Both models showed high hallucination rates on the AA-Omniscience metric at maximum effort. Both models are simultaneously available through Microsoft Foundry, meaning their costs can be applied against existing Azure committed spend. The release follows OpenAI's cancellation of a prior version of the update.
🤖 made with AI
Full details via OpenAI / openai.com/index/introducing…
1
38

