AI News Today (Nov 2, 2026): OpenAI Limits, National Grid & More

AI news today, November 2, 2026: OpenAI is capping paid ChatGPT accounts per device and merging Chat with Work. National Grid is moving ahead with Britain's largest substation for an AI data center, and JPMorgan and Amazon are blocking outside shopping agents. Here are the 10 top stories and 5 security stories, each with free sources.

Key Takeaways

• OpenAI caps paid ChatGPT accounts per device and will merge Chat with Work quotas.
• National Grid agreed to buy 73 acres for a substation serving a Teesside AI data center.
• JPMorgan, Amazon and other retailers are blocking outside AI shopping agents.
• Santander says AI generated €194 million in business value this year.
• A CVSS 10 LXD flaw lets attackers impersonate trusted clients.

1. Update: OpenAI Caps Paid Accounts and Merges Chat With Work

OpenAI now caps the number of paid ChatGPT accounts on a single device and payment card. Users who hit the limit see "Device limit reached," and logging out does not free a slot. The limits cover Go, Plus and the three Pro tiers.

OpenAI is also testing a combined Chat and Work experience and will make ChatGPT and Codex draw from one usage allowance. The change follows the doubled Pro limits OpenAI announced on October 29, which we covered on October 31. The company points users to buying extra credits or waiting for a usage reset instead of opening another subscription.

Why it matters: OpenAI is trading flat-rate simplicity for tighter revenue control, a shift that will test how much convenience subscribers will pay for.

• Source: OpenAI is steering ChatGPT users toward more paid usage — The Decoder
• Source: OpenAI says it's doubling Pro usage limits and bringing Work features to Chat — BleepingComputer

2. National Grid Plans UK's Largest Substation for AI Data Center

National Grid agreed to buy 73 acres at Dorman Point on Teesside for a substation reported to be Britain's largest, supporting a planned AI data center next door. The deal was first reported by Sky News.

The site could host up to 3.2 gigawatts of capacity, Tees Valley mayor Ben Houchen said. Redcar and Cleveland Borough Council has received an outline application for a 400 kV substation, and no decision has been posted. National Grid says the project is part of its Great Grid Upgrade, the largest upgrade of the electricity grid in generations.

Why it matters: AI's power demand is now shaping the grid itself, and Teesside shows how one data center can justify national-scale electricity infrastructure.

• Source: National Grid to site biggest substation on Teesside, near planned AI datacentre — The Guardian
• Source: National Grid sparks deal to pave way for giant UK data centre — This is the Coast / Sky News
• Source: Redcar & Cleveland planning notice, R/2026/0711/OOM — Public Notice Portal

3. Adobe's EditAgent Beats General Image Editors With Specialized Tools

Adobe Research and UC San Diego built EditAgent, an image-editing agent that splits a request into steps and pairs a general editor with specialized tools for sharpening, rotation and zooming. The system outperformed seven other AI agents across three benchmarks.

In 600 edits blind-rated by five professionals, the general editor failed 42.7% of the time. EditAgent generated parallel candidates and used a standard language model to judge them with close-ups and measured statistics, beating fine-tuned evaluators. It needs about 10 times the compute. The verified results also produced about 26,000 training examples for faster editors.

Why it matters: Agentic editing is moving from demos to tool routing and verification, the unglamorous work that makes AI output reliable enough for professionals.

• Source: Adobe is testing how well AI agents can use editing tools like a Photoshop professional would — The Decoder
• Source: EditAgent: Rethinking Agentic Image Editing — arXiv

4. JPMorgan and Amazon Block AI Shopping Agents Like Meta Muse

JPMorgan Chase, Amazon, Charles Schwab, Kohl's and Neiman Marcus are blocking outside AI agents like Meta's Muse from acting for customers, citing security and platform control. JPMorgan CEO Jamie Dimon has said the bank will not let just any agent inside. Amazon says third-party apps should respect service-provider decisions and has asked Meta to remove Amazon from Muse.

A Billions Network audit of 51 large U.S. retailers found none has a verified route for an agent to complete a purchase, and 37 block agents accidentally through old bot rules. Accenture research found 71% of consumers would welcome an AI assistant inside their bank's app.

Why it matters: Banks and retailers are drawing a line around the customer relationship, but the demand data suggests blocking agents may be a holding action, not a strategy.

• Source: For JPMorgan, Amazon, fighting AI bots like Meta Muse may not be a winning long-term strategy — CNBC
• Source: How top US retailers are responding to AI shopping agents — Billions Network

5. Santander and Novartis Point to Tangible AI Gains

Santander said AI generated €194 million ($216 million) of business value in the first nine months of 2026. Novartis said AI had begun to improve its profit and loss on a tangible basis. Both reported third-quarter results this week.

Santander's finance chief linked the gain to lower costs and better customer service. Novartis reported core operating income of $6.04 billion, up 10%, but did not say how much came from AI. Carlsberg said its AI benefits are not yet at scale. Standard Chartered also said AI is helping it grow income faster than costs.

Why it matters: After two years of spending promises, investors now want AI's contribution in the numbers, and only a few companies are starting to show it.

• Source: Companies promised AI gains — here's how the payoff showed up in earnings — CNBC
• Source: Santander Q3 2026 results — Santander

6. Apple CEO Says Apple Won't Ship AI for AI's Sake

Apple CEO John Ternus said the company will keep adding AI to its products but will not ship the technology for its own sake. In an interview with Wired's Steven Levy, he said good AI should be invisible to users.

Ternus, who became CEO in September, said Apple is "absolutely going to leverage AI" but still sees product creation as deeply human. He declined to take a clear position on AI regulation, saying Apple's focus is security and user data. Apple has been criticized for falling behind on AI, especially after delays to a more powerful Siri.

Why it matters: Apple is betting that polished, invisible AI can beat feature-counting, a strategy that will be tested as agents threaten the app model.

• Source: Apple CEO says the company doesn't 'want to ship AI for AI's sake' — TechCrunch
• Source: I Spent 30 Minutes With Apple's CEO. Now I Know Why John Ternus Got the Job — Wired

7. Anthropic Blocks Claude From Recreating Specific Games

Anthropic's Claude Code binary now contains an instruction telling the model not to recreate a specific existing game, even if it is renamed or presented as original. XDA independently found the text in version 2.1.314.

Developers report refusals on game ports and reverse-engineering tasks, though the restriction appears narrower than some early claims. XDA found Opus 5.5 still agrees to decompile a game from a user's own copy. The instruction is delivered remotely, so Anthropic can change it without shipping a new app update. Anthropic has not published a formal policy.

Why it matters: Coding agents are becoming gatekeepers for IP, and the line between a clone, a port and legitimate reverse engineering is being drawn inside the model prompt.

• Source: Anthropic's Claude AI blocks video game cloning and reverse-engineering — Tom's Hardware
• Source: Anthropic seems to be testing a potential crackdown on vibe-coded ports of existing games — XDA

8. Neocloud Debt Fears Grow as AI Bubble Questions Mount

The data-center companies that lease AI computing power are under pressure. CoreWeave carries $35.1 billion in debt, and shares in CoreWeave, Nebius, IREN and Applied Digital are down about a quarter in a month. Gartner expects 80% of neoclouds to cease independent operations by 2030.

The firms borrow against long-term contracts with AI labs and the chips they buy. Microsoft cancelled a deal with Nscale, and Australia's Firmus shelved its IPO. Treasury Secretary Scott Bessent called neoclouds a small part of the AI ecosystem. The concerns have grown as AI labs and hyperscalers cut or delay spending plans.

Why it matters: Neoclouds are the most leveraged layer of the AI buildout, so their credit stress is the clearest early warning that spending may be outrunning returns.

• Source: These firms you've never heard of are feeding fears the AI bubble could burst — BBC
• Source: What's behind the sell-off in AI neocloud stocks like CoreWeave and Nebius — CNBC

9. Indigenous Artist Warns AI Is Stealing First Nations Art

Pitjantjatjara artist Sally Scales said AI is taking First Nations art and turning it into "Temu slop," stripping out the artist's agency, family lineage and lived experience. Scales chairs the National Gallery of Australia's First Nations advisory group.

The National Association for the Visual Arts told a federal AI inquiry that generative models reproduce Aboriginal-style motifs without consent and compete with First Nations artists. It said a general copyright opt-out cannot protect Indigenous cultural and intellectual property. NAVA wants artists to opt in to AI training, rather than having to opt out after their work is used.

Why it matters: Copyright law struggles with communal knowledge, and AI scraping makes the gap between individual authorship and Indigenous cultural ownership harder to ignore.

• Source: AI is stealing Indigenous Australian art and churning out 'Temu slop', artist says — The Guardian
• Source: Artists' rights are in Australia's interests — NAVA

10. 50 AI Agents Fail to Build a Thriving Minecraft Civilization

Y Combinator and Olam Labs put 50 GPT-6.1 Sol agents in a survival Minecraft world for three hours with one goal: build a thriving civilization. They had a shared sandbox, an agent directory and direct messages. They harvested trees, formed a government and sent about 5,900 messages.

A first-night wave of 22 deaths made the agents risk-averse. They built ration ledgers, permits and food queues, then spent much of the run managing scarcity instead of expanding. The experiment is part of a push to study multi-agent behavior.

Why it matters: Agent swarms can coordinate, but the Minecraft run suggests they can also invent bureaucracy and optimize for survival rather than the stated goal.

• Source: AI agent swarms unleashed in Minecraft to 'build a thriving civilization' — Tom's Hardware
• Source: Agents in Games: Civ50 — Y Combinator

AI Security: 5 Stories to Know Today

11. CVSS 10 LXD Flaw Lets Attackers Impersonate Trusted Clients

Canonical fixed a maximum-severity flaw in LXD and MicroCloud that let an attacker authenticate as a trusted client without its private key. The attacker needed network access to the HTTPS API and the client's public certificate.

The flaw is tracked as CVE-2026-112816 and rated CVSS 10.0. Canonical released fixed snap versions for LXD and MicroCloud. There is no sign of exploitation, and Canonical's advisory offers no workaround beyond restricting access to the API. Canonical lists Ubuntu's packages as not affected or not shipped, so the fix is in the snaps.

Why it matters: LXD and MicroCloud underpin private clouds and AI lab infrastructure, so impersonating a trusted client can expose entire clusters and their virtual machines.

• Source: CVSS 10.0 LXD, MicroCloud flaw lets attackers impersonate trusted clients without a private key — The Hacker News
• Source: CVE-2026-112816 — Canonical

12. Polymarket Refunds $2.7M After Supply-Chain Script Theft

Malicious code on Polymarket's website sold signed-in traders' positions and moved about $2.7 million from 726 wallets on October 29. The code replaced a page-speed script the site loaded from DebugBear.

Polymarket says it is returning $3.9 million in total, including $1.2 million for pause-related losses, and has reset sessions and tightened third-party script controls. DebugBear said its DNS records were compromised through its domain registrar. The attackers moved about $2.64 million in POL tokens to their wallet in under an hour.

Why it matters: A single third-party script can inherit a trading site's trust, and the loss shows why pinned versions and subresource integrity matter for crypto and AI services alike.

• Source: Hackers steal about $2.7 million from Polymarket traders via hijacked DebugBear script — The Hacker News
• Source: Polymarket status: October 29 security incident — Polymarket

13. FBI Links North Korea to $6.43B Binance Crypto Heist

The FBI attributed the theft of more than $6 billion in crypto from Binance to TraderTraitor, a North Korean group also tracked as Jade Sleet and UNC4899. The heist is the largest cryptocurrency theft on record.

The FBI, Secret Service and IRS Criminal Investigation said the group is converting the assets mainly to Ethereum and published wallet addresses for exchanges and DeFi services to block. Of the $6.43 billion stolen, $2.85 billion has been frozen on-chain by stablecoin issuers and networks. Researchers say TraderTraitor uses AI-generated lures and fake recruiter personas.

Why it matters: Nation-state crypto theft now funds a broader AI-enabled playbook of fake recruiters and poisoned developer tools, making every hiring inbox part of the attack surface.

• Source: FBI links North Korea's TraderTraitor to record $6.43 billion Binance crypto heist — The Hacker News
• Source: North Korea responsible for $6 billion+ Binance attack — FBI

14. Update: Alpine Linux Rebuilds GitLab After Attempted RCE Attack

Alpine Linux rebuilt its GitLab server from scratch and reset all passwords after attackers tried to exploit CVE-2026-97229, a maximum-severity remote code execution flaw, on October 20 and 21.

Alpine found no evidence that code ran or packages were tampered with, but said missing logs meant it could not rule out a compromise. The attackers downloaded Alpine's GitLab container image, suggesting they were tuning the exploit to its build. The flaw is the GitLab RCE that CISA flagged for active exploitation, which we covered on October 31.

Why it matters: GitLab is where open-source projects build and sign the software that ends up in containers, so an unconfirmed compromise is enough to force a full rebuild.

• Source: Alpine Linux rebuilds GitLab after attempted RCE attack — BleepingComputer
• Source: GitLab security incident: attempted exploitation of CVE-2026-97229 — Alpine Linux

15. Critical RCE Hits AI Project Manager WordPress Plugin

Patchstack disclosed an unauthenticated remote code execution flaw in the Project Manager – AI Powered Project Management plugin for WordPress, affecting versions through 4.1.1. The flaw is tracked as CVE-2026-116512 and rated CVSS 9.8.

Attackers can run commands on the server without logging in. Patchstack says the flaw is expected to be mass-exploited and has issued a mitigation rule; the fix is version 4.1.2. Patchstack says vulnerabilities like this are used in campaigns against thousands of sites at a time.

Why it matters: AI plugins are being added to public websites faster than their security is reviewed, and an unauthenticated RCE gives attackers the whole server.

• Source: WordPress WP Project Manager plugin <= 4.1.1 is vulnerable to a high priority RCE — Patchstack
• Source: CVE-2026-116512 — Tenable

More AI Stories Worth Reading

• Meta's AI ad business is growing faster than Google search, the widest gap since 2021. — CNBC
• Chatbots are giving hikers bad route advice, and search-and-rescue teams are seeing the consequences. — Wired
• OpenAI will train on Daybreak Blue cybersecurity session data from November 29, even for some users who opted out. — The Decoder

Related Reading on Kill The AI

• OpenAI cut its funding target, AI labs backed a scraping opt-out, and a Claude agent injected a mouse brain.
• Google's satellite ran Gemma 4 in orbit, AWS grew 39%, and AI stocks sold off.
• Mistral released Large 4, Microsoft beat estimates, and Nvidia faced antitrust questions over Hugging Face.
• Our data review of the AI boom's first eight weeks, from capex and chips to layoffs.

Methodology & Sources

Compiled November 2, 2026, using multi-source research. Linked outlets: The Decoder, BleepingComputer, The Guardian, This is the Coast, Public Notice Portal, arXiv, CNBC, Billions Network, Santander, TechCrunch and Wired. Also linked: Tom's Hardware, XDA, BBC, NAVA, Y Combinator, The Hacker News, Canonical, Polymarket, FBI, Alpine Linux, Patchstack and Tenable. All sources are free; no paywalls. Details are as reported at compilation time.

Frequently asked questions