AI News Today (Nov 12, 2026): OpenAI $10B, TSMC AI Bet

AI news today, November 12, 2026: OpenAI's quarterly revenue crossed $10 billion, TSMC approved up to $1 billion in Anthropic and OpenAI investments, and Google released AlphaGenome 2. Here are the 10 top stories and 5 security stories, each with free, non-paywalled sources.

Key Takeaways

• OpenAI's quarterly revenue crossed $10 billion for the first time, up almost 50% from Q2.
• TSMC's board approved up to $1 billion in investments in Anthropic and OpenAI.
• Google's AlphaGenome 2 predicts how DNA variants affect specific cell types.
• England's universities regulator urged a return to traditional exams and in-person presentations.
• Railway said stolen AI keys led to about $740,000 in unauthorized charges.

1. OpenAI Quarterly Revenue Tops $10 Billion

OpenAI's quarterly revenue crossed $10 billion for the first time, up almost 50% from $6.7 billion in the second quarter, according to reports. Enterprise sales drove the increase, and ChatGPT now has 100 million paying subscribers, The Decoder reports.

The growth came with a $10.9 billion adjusted operating loss, and OpenAI still trails Anthropic, which reported $17 billion in revenue over the same period. OpenAI is cutting prices and adding enterprise products as it prepares for an IPO next year.

Why it matters: OpenAI is growing faster again, but the gap between revenue and compute spending remains the central question for its public-market debut.

• Source: OpenAI's quarterly revenue reportedly crosses $10B for the first time — TechCrunch
• Source: Enterprise sales push OpenAI's quarterly revenue to $10 billion, though it still trails Anthropic — The Decoder

2. TSMC Bets Up to $1 Billion on Anthropic and OpenAI

TSMC's board approved up to $1 billion in AI investments: $500 million for Anthropic common shares at its IPO price and $500 million for OpenAI preferred stock. The chipmaker said both investments remain subject to the terms of the offerings.

TSMC is the world's largest contract chipmaker. The board also raised its quarterly cash dividend to a record NT$9 per share. The investment gives TSMC direct stakes in two leading AI developers.

Why it matters: The investment ties the world's largest chipmaker more closely to the AI labs driving its growth, adding another link between chip supplier and model developer.

• Source: TSMC to invest in Anthropic, OpenAI, raise quarterly dividend — Focus Taiwan
• Source: TSMC plans up to $1bn investment in Anthropic, OpenAI — Investment Monitor

3. AlphaGenome 2 Predicts DNA Effects by Cell Type

Google DeepMind released AlphaGenome 2, a model that predicts how DNA variants affect specific human cell types and states. It was trained on more than 100 million cell profiles across 68 tissues and identifies nearly twice as many expression-changing variants as its predecessor.

The model is available for non-commercial research through the AlphaGenome API, with a private preview for commercial use. DeepMind says it can reveal disease-relevant effects in cells that broad tissue models miss, including cell states the model has not seen before.

Why it matters: Finer-grained predictions could help researchers prioritize genetic variants for lab work and drug discovery, where cell type often determines whether a mutation matters.

• Source: AlphaGenome 2: Deciphering how cells read DNA to better understand disease — Google DeepMind
• Source: Google's AlphaGenome 2 zooms in on the cellular roots of disease — The Verge

4. UK Regulator Tells Universities to Bring Back In-Person Exams

England's Office for Students told universities to return to traditional exams and in-person presentations after an investigation found AI made coursework "close to impossible" to assess. A survey of more than 300 respondents at 127 institutions found students used AI heavily and often undetected.

The regulator warned that inadequate assessment design could breach registration conditions and lead to sanctions. It said some students could not recall basic details of work they had just submitted, while academics reported batches of near-identical essays with the same sources and style.

Why it matters: Universities are being pushed back toward supervised assessment, a reversal that could reshape how thousands of courses grade writing and research.

• Source: AI has made it 'close to impossible' to assess university coursework, regulator says — The Guardian
• Source: UK regulator threatens universities with sanctions unless they adopt AI-proof exam formats — The Decoder

5. AI Data Centres Plan 181GW of On-Site Gas Power

BloombergNEF tracked 181GW of announced gas-fired generation proposed alongside data centers, more than 90% of it in the United States. Developers are turning to on-site plants because grid connections are taking too long.

The Guardian reports that the planned capacity is almost four times Britain's peak electricity demand. It could emit about 345 million tonnes of carbon dioxide equivalent a year in the US. BNEF expects data center electricity demand to nearly quadruple to 2,354 terawatt-hours by 2035.

Why it matters: AI's power demand is now shaping the energy system directly, and on-site gas plants could lock in decades of emissions long after the current chip cycle ends.

• Source: AI datacentres plan on-site power stations on huge scale despite climate risks, research reveals — The Guardian
• Source: Data Center Power Demand Is Set to Surge, Raising New Climate Challenges — BloombergNEF

6. New York Opens RAISE Act Registration for AI Labs

New York opened a registration portal under its RAISE Act, which requires large frontier AI developers to register with the state by January 1, 2027. Companies with more than $500 million in annual revenue must publish safety frameworks and model transparency reports.

They must also report critical safety incidents within 72 hours. The new DIGIT office will enforce the law, with penalties of up to $1 million for a first violation and $3 million for later ones. The law takes effect at the start of 2027.

Why it matters: New York is building the first state-level compliance regime for frontier AI, and its reporting rules could become a template for other states.

• Source: OpenAI, Anthropic face new oversight as New York's AI safety law takes effect — TechCrunch
• Source: AI Safety: Governor Hochul Announces Next Steps to Regulate Major AI Developers — New York Governor's Office

7. Helsing Unveils Clarity AI Targeting Platform

German defense company Helsing unveiled Clarity, an agentic intelligence platform that uses hundreds of AI agents to analyze satellite imagery, drone video, signals and written reports. It identifies potential targets and prepares data for the weapons system best suited to an engagement.

Helsing says operators review the evidence, choose which assets to engage and approve every transfer to a weapon, with each decision logged. The system runs on customer-owned infrastructure and is built in Europe, positioning it as a sovereign alternative to Palantir's Maven.

Why it matters: The platform brings agentic AI deeper into military targeting, where the speed of machine analysis can outpace the human review it is meant to preserve.

• Source: Helsing launches European alternative to Palantir's military AI — Kyiv Independent
• Source: Helsing introduces Clarity AI platform to identify potential military targets for operator review — Defence Industry Europe

8. Big Walk AI Use Sparks Indie Game Reckoning

Nineteen members of the Independent Games Festival jury said they would not vote for any game that uses generative AI "in any capacity." House House, the developer of indie hit Big Walk, then disclosed that LLMs helped with under 0.1% of the game's code.

House House said it used chatbots mainly to research third-party code issues and fix a shadow bug. It disclosed the use in its IGF submission but not on Big Walk's Steam page, saying the platform does not ask about code or chatbot usage.

Why it matters: The dispute shows how a tiny amount of AI assistance can disqualify a game from a major award, forcing studios to disclose tools they treated as routine.

• Source: The indie game community is in disarray over AI and Big Walk — The Verge
• Source: House House Discloses AI Use For 'Under 0.1%' Of Big Walk's Code Following IGF Voting Controversy — Game Informer

9. CoreWeave Shares Jump 26% on Q3 Beat

CoreWeave shares jumped 26% after the AI cloud provider reported $3.66 billion in third-quarter revenue and raised its full-year guidance to $13.0–$13.3 billion. CEO Michael Intrator said demand for AI infrastructure remains "overwhelming."

The company said it is oversubscribed for 2027 capacity and expects $20–$21 billion in annualized run-rate revenue by the end of 2026. It also raised its capital expenditure forecast to $37–$39 billion.

Why it matters: CoreWeave's results eased fears about the AI trade for a day, but its rising capex and debt load show how much cash the buildout still consumes.

• Source: CoreWeave soars 26% as CEO reaffirms AI compute demand after strong Q3 report — CNBC
• Source: CoreWeave Reports Strong Third Quarter 2026 Results — CoreWeave

10. US Data Brokers Sell $500M in RL Environments to Chinese Labs

A ChinaTalk investigation found at least nine US data vendors, including Mercor, Surge AI, AfterQuery and Turing, selling roughly $500 million in training data to Chinese AI labs. Some of that data is reinforcement-learning environments previously sold to OpenAI or Anthropic.

Chinese labs reuse the same environments to cut compute costs and skip costly trial and error. The report says the trade is largely unregulated and calls for export controls, especially for environments used in cybersecurity and biology.

Why it matters: RL environments encode how models learn to perform tasks, and selling the same ones to rival labs blurs the line between commercial data and strategic technology transfer.

• Source: Piggybacking to the Frontier: The Firms Selling RL Environments to Chinese Labs — ChinaTalk
• Source: US data brokers are selling half a billion dollars in RL training environments to Chinese AI labs — The Decoder

AI Security: 5 Stories to Know Today

11. Railway Breach Leads to $740,000 in Fraudulent AI Charges

Railway said attackers exploited a previously unknown flaw in its build infrastructure on October 20, escaped filesystem isolation and reached systems storing customer environment variables. The company disclosed the breach on October 22.

Secrets in at most 1.85% of workspaces, about 100,000, were exposed, and stolen AI provider keys generated an estimated $740,000 in unauthorized usage. Railway says it will pay those charges when an AI provider will not refund them. It reset linked-account tokens for 1,852 users.

Why it matters: AI API keys are bearer tokens for expensive compute, so a platform breach can turn into a large bill before anyone notices.

• Source: Railway Says Stolen Customer AI Keys Led to $740,000 in Estimated Unauthorized Usage Charges — The Hacker News
• Source: Railway breach leads to $740,000 in fraudulent AI charges — BleepingComputer

12. AntiFox AI Pentesting Platform Used in AWS Theft

Arctic Wolf Labs tied a series of intrusions to AntiFox, an open-source agentic penetration-testing platform hosted on Alibaba Cloud. Attackers used it to exploit internet-facing web services, deploy web shells, validate credentials and enumerate AWS environments.

In some cases they obtained production secrets. The attackers used valid IAM credentials and EC2 roles, cycling through identities and making hundreds of AWS API requests in about 30 minutes. AntiFox is one of about 45 open-source agentic attack frameworks.

Why it matters: Agentic pentest tools lower the skill and time needed to turn a web foothold into cloud credential theft, and defenders must now match the agent's speed.

• Source: AntiFox AI Pentesting Platform Linked to Attacks That Accessed AWS Production Secrets — The Hacker News
• Source: Hackers use AntiFox AI agents in AWS credential-theft attacks — BleepingComputer

13. Over 35,000 Grok-Built Apps Exposed Without Login

Claroc found 37,571 Grok-built apps online and 35,578 of them open to anyone without a sign-in. The exposed apps included 24 built by X and SpaceXSI staff. They contained a confidential onboarding plan, a sales territory map of 594 accounts and an ad proposal for a federal agency.

About one in 10 open apps contained internal business or personal data. Claroc found a hospital group's searchable list of more than 3,500 employees, an intensive-care briefing on a patient and a contractor's timesheet with every employee password in plain text. SpaceXSI removed the 24 staff apps but not other users' apps.

Why it matters: AI app builders can turn a chat prompt into a public website in minutes, and the default sharing setting is now a data-loss channel.

• Source: Over 35,000 Grok-Built Apps Exposed Online, Some Leaking Confidential Data and Passwords — The Hacker News
• Source: Grok-built apps exposed patient data, plaintext passwords, and SpaceXSI's own sales pipeline — The Decoder

14. Microsoft Lazarus Backdoor Hides in Fewer Than 10 Neurons

Microsoft researchers showed that editing fewer than 10 rarely used neurons can implant a hidden backdoor in an AI model without retraining it. The model behaves normally until a trigger appears, then outputs an attacker's command that an agent may run with its existing permissions.

The technique, called Lazarus, can be activated by something as ordinary as a date change. Microsoft says it has not seen threat actors use it, but warns that anyone who can modify model weights through a compromised account, pipeline or storage system could plant it. Standard safety tests may miss it.

Why it matters: Agent security depends on model integrity, and a backdoor that survives evaluation turns a trusted model into a supply-chain risk.

• Source: Microsoft finds AI agents can be backdoored by editing under 10 neurons — The Next Web
• Source: Microsoft researchers demonstrate hidden backdoor technique for AI agents — Help Net Security

15. SGLang RCE Breach Exposes NEAR AI Agents and Users

An attacker exploited a remote code execution flaw in SGLang, an open-source inference engine, and ran code on one of NEAR AI's inference servers last Sunday. Because SGLang ran inside a trusted execution environment, the attacker could see requests processed there, including some sent in confidential mode.

NEAR AI confirmed that account details for about 100 users were downloaded. It also said 18 AI agents on an older deployment were accessed, exposing chat history, settings, logs and two API keys. The company revoked sessions and credentials, replaced agent tokens and fixed the SGLang flaw on its servers.

Why it matters: Confidential computing protects data from the host, but not from a vulnerability in the software running inside the enclave, and inference engines are now high-value targets.

• Source: SGLang RCE Exploit Breaches NEAR AI, Steals Data From 18 Agents and About 100 Users — The Hacker News
• Source: SGLang RCE Breach at NEAR AI: 18 Agents and 100 Users Compromised — Aviatrix

More AI Stories Worth Reading

• Oura will acquire AI-native primary care startup Counsel Health for $625 million in cash. — TechCrunch
• eBay appears to have blocked third-party AI agents, including Meta's Muse, from browsing its listings. — The Verge
• Google Labs launched Tabletop, a waitlisted experiment for building shared trip itineraries, polls and brackets. — Google AI Blog

Related Reading on Kill The AI

• Anthropic's IPO terms, Nvidia's Wachtell probe, and Gemini 4 Carbon.
• Nvidia's board probe, Claude Fable 5.2, and Meta's Muse Ember.
• Nvidia's spy inquiry, Apple's agentic Siri, and OpenAI's $11 billion raise.
• Our data review of the AI boom's first eight weeks, from capex to layoffs.

Methodology & Sources

Compiled November 12, 2026, using multi-source research. Outlets: TechCrunch, The Decoder, Focus Taiwan, Investment Monitor, Google DeepMind, Google AI Blog, The Verge, The Guardian, BloombergNEF and New York Governor's Office. Also Kyiv Independent, Defence Industry Europe, Game Informer, CNBC, CoreWeave, ChinaTalk, The Hacker News, BleepingComputer, The Next Web, Help Net Security and Aviatrix. All sources are free to read; no paywalled links. Details are as reported at compilation time.

Frequently asked questions