Read
Latest news
The day-to-day of tech, straight from the newsroom.
53 news stories · Updated 10 hr. ago
Postman installs Codex and Claude plugins without notice and approves its own hooks
An issue opened in Postman's official repository shows the desktop app installing AI plugins in agents like Codex and Claude Code without asking for consent, and approving on its own the hook it created itself.
Bluetooth Core 7.0 adopts HDT and boosts BLE throughput by nearly 4x
The gain targets high-resolution audio, but directly affects wearables, IoT, and personal AI devices.
US Voters Reject Data Centers in More Than 30 Cities at Tuesday's Polls
In at least nine US states, residents approved moratoriums and bans on data centers in Tuesday's election (November 3), a direct message to the accelerated expansion of AI infrastructure.
Fake North Korean job interview led to $6.3 billion theft at Binance
A Binance report published this Tuesday details how a months-long campaign, with fake interviews and malware disguised as a technical test, paved the way for the largest cryptocurrency theft on record.
Oracle Cancels $7 Billion Deal With Tencent Under Trump Pressure, FT Says
Oracle backed out of leasing AI chips to Tencent after the US government intervened, according to the Financial Times. The case shows how political decisions can unravel cloud contracts signed between private companies.
Apple opens Siri AI to any developer to integrate apps via App Intents and skills
In the iOS 27.3 Developer Preview, released on November 3, Apple opens up the App Intents framework to any app and introduces skills, which teach the assistant to chain several actions into a single request.
GitHub Suffers Two-Hour Outage in Pull Requests, Git Operations, and Issues
A failure in GitHub's infrastructure degraded Git Operations, Pull Requests, Issues, API Requests, and Actions for about two hours this Wednesday morning (Nov. 4). The service was restored, but the root cause has not yet been disclosed.
CrowdStrike pays nearly $500 million for two AI agent security startups
The cybersecurity company closes the purchase of Fiddler AI and Dreadnode to monitor and test AI agents before and after deployment, in a market growing alongside the uncontrolled adoption of AI inside companies.
Amadeus Breach Exposes Data From 46,000 Singapore Airlines Bookings
Singapore Airlines confirmed that a security breach at Amadeus, its booking technology provider, affected around 46,000 records. The case exposes the risk of relying on a single platform used by airlines worldwide.
Raspberry Pi 5 Price Rises for the Fourth Time in 2026, Now Triple Its Launch Price
Raspberry Pi Holdings has raised prices four times in 2026 alone. The 16 GB Raspberry Pi 5, which cost $120 in 2023, now reaches $385 in the United States.
Marvell Buys Lumentum for $115 Billion in Optical Chip Consolidation for AI
Marvell announced the purchase of Lumentum, a maker of optical chips, in a deal valued at up to $115 billion. It is the company's fourth acquisition in photonics since December 2025, a move that concentrates in few hands the technology that connects AI data centers.
DeepSeek may train V5 model on Huawei's Ascend chips, Chinese newspaper says
A report by the Chinese newspaper STAR Market Daily, published on November 3, says DeepSeek is considering using Huawei's Ascend processors to train the V5 model, not just to run inference as it does today.
TorchTPU: Google and Meta Open Native PyTorch Backend for TPUs
New open source project eliminates the code rewrite that was blocking TPU adoption for PyTorch developers, with tests on 5,000 Hugging Face models and benchmarks on TPUv7 chips.
IT stopped being a shortcut to the good life, says viral post on Habr
A post by Russian developer @1endstick, viral on Habr since November 1st, argues that AI won't mass-lay off programmers, but will level salaries and hollow out exactly the comfortable mid-career range in IT.
Uber Builds MCP Gateway to Govern 5,000 AI Agent Tools
The company created a central layer for discovery, authorization, and execution to keep AI agent adoption from turning into a mess of duplicated integrations and loose permissions.
Research shows poisoning 0.1% of pretraining data is enough to implant a backdoor in AI agents
Study published in October 2026 shows that contaminating a minimal fraction of the pretraining corpus is enough to plant a remote-execution backdoor in LLM-based command-line agents.
Lumi launches with 4 billion parameters and runs without a data center GPU
Lumi is the language model released by the Centro de Excelência em Inteligência Artificial at Universidade Federal de Goiás.
IaC and tested backups prevented data loss when Yandex data centers went down
In October 2026, attacks on three Yandex data centers took down the infrastructure of a client of KTS, a Russian DevOps consultancy. Infrastructure as code, living documentation, and backups tested in advance kept the team from rebuilding from scratch.
Fortanix integrates hardware attestation into Kubernetes' Confidential Containers
Fortanix connected its attestation and key management service to the CNCF's Confidential Containers project, releasing secrets only to verified pods inside hardware enclaves. For teams running AI models on shared Kubernetes, this takes the job of building that trust layer from scratch off their plate.
DeepSeek releases dsh-superagent code, an agent that opens pull requests on its own
DeepSeek's Harness team has open-sourced dsh-superagent, a research prototype that takes development requests via GitHub (and optionally Feishu) and returns ready pull requests. The project is MIT-licensed and grew out of the team's own internal workflow.
Trigger.dev suffers security incident, exposes customer environment variables
Trigger.dev confirmed this Sunday (November 1) that an attacker accessed environment variables belonging to a subset of customer organizations; the company says it has blocked the flaw and notified affected users by email.
Meituan launches LongCat-2.5, a 1.6-trillion-parameter model for coding agents
China's Meituan unveiled LongCat-2.5 on November 2, a multimodal model with 1.6 trillion parameters trained to sustain long coding tasks with AI agents.
DoorDash Cuts Grocery Reorder Latency From 16 to 2 Seconds With AI Agents
The company now pre-computes with AI the items customers typically reorder, instead of recalculating everything with every order. The result: less waiting, more completed orders.
Lean creator says AI made formal verification viable for everyday dev work
In an interview with The Data Exchange podcast, Leo de Moura, creator of Lean and VP at AWS, explains how AI models now translate code from any language into machine-checked mathematical proofs, from compression libraries to cryptography libraries.
AWS Has Reserved Capacity Sold Out Through 2027, But Sets Aside a Share for On-Demand Use
Andy Jassy confirms that large blocks of AWS capacity are largely sold out through 2027, with part of 2028 already allocated. The company says it keeps room for on-demand EC2 and Bedrock, but developers are already reporting queues for common instances.
Cloudflare Swaps Attribute-Based Encryption for Post-Quantum Scheme in GeoKey Manager
Cloudflare's geographically restricted key manager entered closed beta with a hybrid post-quantum scheme, replacing the attribute-based encryption used since 2022. The decision to fail closed on verification errors is dividing the technical community.
Cybersecurity Startup Accesses Anthropic's Internal Systems in Under 12 Hours
Cathedral, founded by former DOGE members, used AI agents to find flaws in sales and HR software used by Anthropic, exposing employee data and internal messages. The case reignites the debate over third-party software security and AI agents that bypass access restrictions.
- QCon
QCon AI New York 2026 to discuss agent identity, context cost, and open source inference
The QCon AI New York 2026 program committee selected six sessions on the bottlenecks teams already face when putting agents into production: authorization, context cost, and standards like MCP and A2A.
RAMBO: research shows how to turn RAM into a radio antenna to leak data from air-gapped PCs
An academic paper describes RAMBO, a technique that converts the memory of an air-gapped computer into a radio transmitter capable of leaking files, passwords, and encryption keys at up to 1000 bits per second.
Project compiles the NES's entire CPU into a Qwen model's weights, without training anything
The nesformer project turns the NES's 6502 chip, its RAM, and the cartridge ROM into the weights of a Qwen3.5 model, computed by a compiler, not by training. The result actually runs Super Mario Bros., but hundreds of times slower than the original console.
New Wave of Cyberattacks Hits South Korea's Banks and Brokerages
Daishin, Mirae Asset, and crypto exchange Bithumb were targeted by breaches in October, a month after researchers linked an earlier attack to an AI-based offensive pentesting tool.
Mistral Large 4 (Le Chonk) becomes a 1-trillion-parameter open model
Mistral AI released the weights of Large 4, codenamed Le Chonk, a multimodal model with 1 trillion parameters that leads open code benchmarks, with a high refusal rate for malicious security requests. For those looking to escape closed APIs, the announcement comes with important caveats about infrastructure and licensing.
Malformed WAP Push SMS Crashes Calls, Including Emergency Calls, on Android
A malformed WAP Push message kills Android's telephony process without any interaction from the victim, dropping active calls and leaving the device stuck in a reboot loop.
Google details PageBreak agent that has already found over 500 XSS flaws
In a paper published this week, Google describes PageBreak, a Gemini-based agent that only reports a vulnerability after reproducing it with a real exploit against the running system.
Official Claude Code extension on Cursor Marketplace caught with SSH backdoor
Researchers at Phorion found a persistent SSH backdoor embedded in versions 2.1.312 and 2.1.313 of the Claude Code extension distributed through Cursor Marketplace, for macOS and Windows. Cursor has already removed the packages, but removing the extension is not enough: SSH keys and persistence mechanisms must be cleaned up manually.
Brazil reaches 10.2 million devs on GitHub and becomes the world's 4th largest market
The Octoverse 2026 report, presented at the GitHub Universe conference in San Francisco, confirms Brazil as the world's 4th largest developer community, growing faster than almost anywhere else. The shift comes as AI agents already write the first draft of more than 40% of the code approved on the platform.
CVE-2026-98432: use-after-free flaw in Linux TCP exposes remote root via MPTCP
A use-after-free flaw in the Linux kernel's TCP stack allows remote code execution as root on servers with Multipath TCP enabled. The case shows why relying solely on the kernel version number isn't enough to know whether a server is protected.
Anthropic and METR research shows risk to training new engineers from heavy AI use
Data from METR, Anthropic, and the consultancy DX show that AI tools speed up routine tasks but erode exactly the skills that juniors used to spend years building, skills that would later be needed to oversee the AI-generated code itself.
Microsoft retires Azure Deployment Environments and Dev Box with no equivalent replacement
Azure Deployment Environments shuts down on February 22, 2027, with no replacement product at all. Microsoft Dev Box continues until September 2028, after which users are redirected to Windows 365, which Microsoft itself says is not equivalent.
Quantized world model runs five full games inside a Raspberry Pi Pico 2 W
The tinyworldformer project trains two tiny neural networks that generate and play Breakout, Pong, Flappy Bird, Doom-lite, and Pac-Man lite directly on the RP2350, with no cloud and no GPU.
AI advances in mathematics and raises alarm over the foundation of current cryptography
UK regulators are already testing banks for a scenario in which AI models and quantum computers threaten RSA. In Brazil, banks like Inter and companies like Grupo FS are starting to take notice.
Mistral releases weights of 1-trillion-parameter Large 4 model, aimed at cybersecurity
Mistral released the weights of Large 4, nicknamed Le Chonk, allowing companies and governments to download and run the model on their own infrastructure for cyber defense.
Cloudflare launches Web Search API for AI agents on AI Gateway
Cloudflare launched in beta a web search API integrated into AI Gateway, allowing Workers to query the internet in real time with unified login, access control, and billing.
Anthropic launches tools to measure the return on enterprise AI spending
Anthropic's new reporting package shows managers how much each team spends on tokens, what it produced, and where the workflow got stuck. The launch comes alongside data showing the company's revenue growing 900% among the same customers.
EDK2 vulnerability compromises secure boot for most of the world's servers
CVE-2026-97631, with a severity score of 9.5, lives in the open-source EDK2 used by nearly every UEFI firmware vendor on the market. The fix leaked before the coordinated disclosure, and now any attacker can study the exploit.
Microsoft fixes Windows Shell flaw already exploited in real-world attacks
Of the 1,415 vulnerabilities fixed in the October 2026 Patch Tuesday, one in Windows Shell already has confirmed exploitation. See what changes for those who develop and maintain Windows and Office applications.
Microsoft Discloses Azure's Dollar Revenue for the First Time: $33.2 Billion
Until now, Microsoft only talked about Azure's growth in percentage terms. In the earnings report for the quarter ended in September, the company broke out the dollar figure for the first time, making it possible to compare directly with AWS and Google Cloud.
Two critical PHP flaws expose 70% of the web and trigger mass patch alert
Wiz CTO, the security arm acquired by Google, details in a podcast how teams from OpenAI, PHP, and researchers rushed to close two critical flaws that hit about 70% of the web's sites.
Google opens Gemini and Search to third-party apps with the Connected Apps Platform
The new Connected Apps Platform lets any developer register an app to be triggered inside Gemini and, soon, Google Search. The flow runs through an MCP server, AppFunctions, and a monitoring console, currently in beta only for the United States.
Git on macOS still trusts eMudhra's compromised certificates
More than ten days after the alert, Apple hasn't fixed git, curl, and python3 on macOS: the tools still accept compromised eMudhra TLS certificates, and the system's latest update undid the manual fix developers had applied.
End of the 50 most recent news stories. Subscribe to the RSS feed to follow along in real time.