The AI stories worth your time today.

Photo: Chris Montgomery / Unsplash

SecurityBusinessThe Hacker News ↗13 Nov

FBI says North Korean hackers stole over $400 million in crypto using deepfake video calls

The FBI says hackers tied to North Korea stole more than $400 million in cryptocurrency over 18 months in a social engineering scheme codenamed FaceStealer that used deepfake video calls. The finding was part of a coordinated set of US government actions.

Why it mattersA familiar face on a video call is no longer proof of identity, so crypto firms and finance teams need extra checks before moving money.

EverydayBusinessHelp Net Security ↗13 Nov

OpenAI rolls out ChatGPT Wallet so AI agents can buy things without seeing your card number

OpenAI began rolling out ChatGPT Wallet, which lets its agents pay for purchases using a stand-in card number, so card details are not shared with the agent or the merchant. It was built with American Express, Mastercard and Visa, starts with US Pro subscribers, and requires biometric approval for each purchase.

SecurityBusinessBleepingComputer ↗11 Nov

Breach at AI provider Fireworks exposes some Notion AI chat transcripts

Notion disclosed that a security breach at its AI infrastructure provider, Fireworks AI, exposed Notion AI chat transcripts in fewer than 1% of workspaces. It is Fireworks' second data breach in two months.

More from today

1. SecurityBuildersCSO Online ↗12 Nov

   Hackers stole AI provider keys from Railway and ran up $740,000 in fraudulent usage

   Cloud host Railway confirmed a nearly 26-hour fight against attackers who hunted for AI provider credentials, generating about $740,000 in fraudulent AI charges. The stolen keys were sought for resale to services used for model distillation, and up to 100,000 workspace credentials may have been exposed.

   Why it mattersAI API keys are now valuable loot, so developers should store them securely, rotate them and set spending limits.

   

2. PolicyBusinessThe Decoder ↗13 Nov

   Researcher says model weights show six Chinese AI models were trained on export-banned Nvidia chips

   Former OpenAI researcher Will DePue says he found technical evidence that DeepSeek, Moonshot AI, MiniMax and Ant Group used Nvidia Blackwell GPUs despite US export controls, by reading chip signatures from published model weights. He estimates the hardware behind the training runs cost more than half a billion dollars.

   Why it mattersIf confirmed, the method gives regulators a new way to check whether chip export controls are working, adding pressure on Nvidia during its leadership crisis.

   

3. BusinessThe Decoder ↗13 Nov

   Anthropic IPO is six times oversubscribed and on track to be the largest ever

   Orders for Anthropic's IPO exceed available shares by six times, putting it on track to be the biggest public offering in history. Banks are rejecting buy orders from mainland China and Hong Kong, citing US export control rules.

   Why it mattersHuge investor demand shows confidence in AI labs remains strong, which affects funding and competition across the industry.

   

4. SecurityBuildersThe Hacker News ↗11 Nov

   Over 35,000 apps built with Grok's coding agent found open online, some leaking passwords

   Security firm Claroc found 37,571 apps built with Grok online, and 35,578 of them could be opened without signing in, including 24 built by staff at X and SpaceXSI. About one in ten contained internal business data, and some leaked confidential information and passwords.

   Why it mattersAnyone building apps with AI coding tools should check access settings before publishing, since default setups can expose private data to anyone.

   

5. SecurityBuildersThe Hacker News ↗11 Nov

   Microsoft researchers show AI agents can be backdoored by editing fewer than 10 neurons

   Microsoft researchers showed a technique called Lazarus that plants a hidden backdoor in an AI model by changing fewer than 10 rarely used neurons, with no retraining. When a trigger appears, an agent using the model runs an attacker's command, yet it behaves normally during standard safety tests.

   Why it mattersCompanies that give AI agents access to terminals or systems need to verify where model weights come from and who can change them, because safety tests may not catch this kind of tampering.

   

6. BuildersThe Decoder ↗12 Nov

   Google DeepMind launches AlphaGenome 2 to study genetic variants by single cell type

   Google DeepMind unveiled AlphaGenome 2, which analyzes genetic variants for individual cell types instead of bulk tissue and was trained on more than 107 million cell profiles. It finds nearly twice as many disease-relevant gene changes as the earlier version and is available for noncommercial research through an API.

   Why it mattersResearchers studying genetic diseases get a more precise tool that could help find which cells are affected by a mutation.

   

7. SecurityPolicyDark Reading ↗11 Nov

   Japan pushes emergency cybersecurity funding as AI-powered attacks rise

   Japan is moving forward with emergency cybersecurity funding in response to more AI-driven cyberattacks. The package includes a joint public-private cyber response team, an AI specialist team led by the economy ministry, and wider room to use the government's new active cyber defense powers.

   Why it mattersGovernments are starting to treat AI-assisted hacking as an urgent national threat, which can shape how companies in Japan work with public authorities on defense.

   

8. BuildersThe Decoder ↗14 Nov

   Runway releases Praxis-1, an open-weight robot control model built on video AI

   Runway released its robotics model Praxis-1 with open weights and source code under the Apache 2.0 license. It uses knowledge from video training to control robots and leads its category on Nvidia's RoboLab-120 benchmark, but cannot yet plan complex action sequences on its own.

   Why it mattersRobotics developers get a free, openly licensed model to build on, lowering the cost of experimenting with AI-controlled robots.

   

9. SecurityEverydayDark Reading ↗13 Nov

   ArcticRAT Android malware gang now uses AI to profile victims' finances

   The ArcticRAT cybercrime group has expanded its Android malware from crypto theft to full device control. The attackers have added AI-based analysis of stolen data to profile victims' finances.

   Why it mattersCriminals are using AI to pick the most valuable targets, so mobile banking users should be careful about app installs and permissions.

   

10. BusinessBuildersThe Decoder ↗13 Nov

    Anthropic launches Claude for Manufacturing plugin connecting to factory business software

    Anthropic's Claude for Manufacturing plugin connects to Autodesk Fusion, Oracle NetSuite and Infor, with 17 skills for tasks like drafting quotes, comparing suppliers and placing backup orders after human approval. It does not control physical machines yet.

    Why it mattersManufacturers can start automating procurement and planning paperwork with AI agents while keeping humans in charge of approvals.

    

11. SecurityBuildersHelp Net Security ↗13 Nov

    Pinned MCP server quietly ran with 15 different dependency sets over five weeks

    A Corelight researcher found that an MCP server set to run one fixed release changed its dependencies 15 times in five weeks without any configuration change. Twenty-three of its 69 packages changed versions, including cryptography, JSON Web Token and MCP SDK libraries.

    Why it mattersTeams connecting AI coding assistants through MCP servers should lock down full dependency trees, not just the main version, to avoid unexpected supply chain risks.

    

12. EverydayBusinessThe Decoder ↗13 Nov

    Meta's Muse agent can now phone businesses for paying US subscribers

    Meta's AI agent Muse can now call businesses such as cable providers and restaurants for paying US subscribers, and Meta says the agent has over 10 million users. Similar features exist from Google Gemini, Instinct and xAI, but many businesses hang up once they detect an AI caller.

    Why it mattersAI agents calling on your behalf are becoming common, so businesses must decide how to handle them and consumers should expect mixed results.

    

13. BusinessBuildersThe Decoder ↗14 Nov

    OpenAI breaks up its robotics team as general models take over robot control

    OpenAI is dissolving its standalone robotics team for the second time and moving the researchers to other teams. General-purpose models like GPT-6 Astra can reportedly control robots directly, and the company says the change will speed up its robotics work.

    Why it mattersIt signals that leading labs think one general model may handle robotics too, which could change how robotics startups and investors plan.

    

No stories for this audience today. Try another filter.