US Imposes 100% Tariffs on Sensitive Drones and Their Components

President Trump signed a proclamation imposing tariffs on imported drones and their components, which the White House says is needed because commercial and US military drones “rely on foreign sources for critical Unmanned Aircraft System (UAS) components.” The proclamation sets a 100% ad valorem duty on drones it deems “particularly sensitive for national security purposes” — a category including drones with a maximum takeoff weight above 25 kilograms and drones with thermal imaging — along with their docking stations and certain critical components. A 25% tariff applies to certain smaller drones and other components, 15% to drones and components from the European Union, Japan, Liechtenstein, South Korea, Switzerland, and Taiwan, and 10% to drones from the UK, provided substantially all hardware, software, and technology originates in those countries and the United States.

The tariffs take effect 21 days after signing, which TechRadar dates to September 3, 2026, while tariffs on non-sensitive drone components take effect 180 days after signing. The proclamation authorizes the Commerce Department to establish an onshoring program for new drone-manufacturing investment; TechRadar notes that specialized imported drones used for search and rescue and power-line inspection have no domestic manufacturing capacity to absorb the demand.

Featured source: TechRadar , White House fact sheet .

Other Stories

Anthropic Backers Expect an October IPO Valuing the Company Above $2 Trillion

Half a dozen of Anthropic’s backers told the Financial Times they expect the AI lab to float at a valuation of $2 trillion or more in October, a figure that would eclipse SpaceX and make the debut the largest-ever initial public offering. The investors expect annualized revenue between $100 billion and $120 billion by the end of 2026 on the company’s preferred measure, up more than tenfold over the course of the year. Two investors said a temporary Commerce Department ban on Anthropic’s best models slowed revenue growth in June before a rebound, executives had yet to fix a valuation target, and Anthropic declined to comment.

Filed from: Ars Technica .

Nvidia Guarantees GPU Collateral Value in Up to $500 Billion of AI Data Center Financing

Nvidia announced that Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR are willing to commit up to $500 billion to build AI data centers, with Nvidia using its own money to guarantee that chips used as collateral in the deals retain value, covering up to 25% of the difference if GPUs cannot command the expected resale price. TechCrunch notes the guarantee creates “wrong way” risk, because Nvidia’s obligations would grow as demand weakens, and that bond markets reacted strongly enough for CEO Jensen Huang to explain the limits of the company’s exposure on X and business TV. The arrangement is also designed to seed a secondary market for aging GPUs.

Filed from: TechCrunch .

Judge Orders Google to Remove Friction From Rival App Store Installs on Android

Judge James Donato ordered Google to remove what he called “anticompetitive friction” from installing rival app stores on Android, in the long-running Epic v. Google case. The court directed Google to surface third-party app stores in Play Store search results, drop a “view” step that precedes “install” for third-party app stores, and stop funneling app-store searches through a banner to a dedicated page. The hearing followed a live demonstration by Epic of how many steps rival installations currently require, just over a month after the two sides appeared to stop fighting over the future of Android app distribution.

Filed from: The Verge .

White House Authorizes Private Security Firms to Run Offensive Cyber Operations Abroad

A National Security Presidential Memorandum directs the National Coordination Center, which operates under the Homeland Security Task Force, to build a program in which private security firms conduct “Cyber Surveillance Operations and Cyber Effects Operations” against foreign transnational criminal organizations that hack US persons, organizations, or government entities, with the Justice and Homeland Security departments providing oversight. Ars Technica says the program is the first time the federal government has authorized private companies to conduct offensive cyber operations against overseas hackers, with ransomware, sextortion, phishing, financial fraud, and impersonation scams listed as eligible targets.

Filed from: Ars Technica .

Uber and Pony.ai Plan to Bring 2,000 Robotaxis to Europe

Pony.ai and Uber plan to deploy 2,000 robotaxis across four European cities under an expanded partnership, extending the joint-deployment model first pursued in Zagreb, Croatia with local partner Verne. Pony.ai will supply the autonomous-vehicle technology and Uber the ride-hailing platform, with fleet management possibly handled by local providers and ownership varying by market. The companies did not name the cities or provide a deployment timeline.

Filed from: TechCrunch .

Cloudflare Adds MCP Traffic Detection and Control to Cloudflare One

Cloudflare announced Cloudflare One capabilities that identify and inspect Model Context Protocol (MCP) traffic on managed networks, showing which users and servers generate it, and that combine with its MCP Server Portals so administrators can enforce approved connection paths for AI agents. The company argues that permission designs built for human users do not bound agents, since one nondeterministic decision can become thousands of tool calls, and notes MCP traffic has no guaranteed hostname or path shape, so a direct connection can look like any other HTTPS API call.

Filed from: Cloudflare Blog .

Trezor Confirms 13,000 Customers’ Details Exposed in Logistics Breach

Trezor confirmed that a breach at logistics partner ShipMonk exposed personal data of more than 13,000 customers, including names, email addresses, phone numbers, and shipping addresses of 11,742 customers in the US, UK, Sweden, Colombia, Brazil, Italy, and Portugal who ordered between May 10 and August 8. A further 1,947 customers had names, home cities, and email addresses exposed, some possibly for orders before May 10, a timeframe Trezor said it is verifying with ShipMonk. Trezor called it the first breach since its 2013 founding to expose customer phone numbers and shipping addresses, and warned affected customers of an increased risk of phishing attempts.

Filed from: The Register .

Researchers Find WebKit Leaks That Expose iCloud Private Relay Users’ IP Addresses

Researchers Talal Haj Bakry and Tommy Mysk found three WebKit features — DNS prefetching, WebAuthn Related Origin Requests, and WebTransport — that can bypass proxy configuration in iOS and macOS browsers and send traffic directly from the device, exposing the user’s real IP address and DNS servers, and said the leaks also affect Apple’s iCloud Private Relay. Apple issued a patch after the findings circulated, TechRadar reported, though it says Apple was reportedly aware of the underlying flaw for at least a year. The researchers fixed the leaks in their own Psylo browser, version 1.3.1.

Filed from: TechRadar , Mysk research .

Experts Warn Autonomous AI Attacks Pose ‘Clear and Present Danger’ to Critical Infrastructure

The Register reports that attackers in early July used an attack framework built on the open-source Hermes and OpenClaw AI agents to break into a Taiwanese government website and ultimately compromise a government email system, the country’s nuclear safety agency, IT supply-chain vendors, and at least seven energy companies, with the “near-autonomous” system deploying up to eight sub-agents across 12 “attack waves.” Citing officials and vendors at Black Hat, it quotes TrendAI’s Tom Kellermann calling autonomous AI attacks on critical infrastructure a “clear and present danger,” and an FBI Cyber Division assistant director saying the bureau is focused on the downstream, potentially kinetic impact of targeting critical infrastructure.

Filed from: The Register .

X Open Sources Its For You Ranking Algorithm and Adds Account-Labels Transparency

X said it is making the source code for its “For You” timeline and core ranking engine available on GitHub under the Apache v2 license, expanding its open-source codebase roughly 10 to 15 times to include model configuration, filters, and the parameters used to weight ranking signals. The company is also adding a feature that lets users see whether their account or posts have been impacted by its ranking systems.

Filed from: TechCrunch .

Microsoft Merges Its Copilot Apps and Retires Several AI Features

Microsoft is combining its consumer Copilot app and the Microsoft 365 Copilot app into one product and retiring a number of unsuccessful AI features, as initially reported by GeekWire and detailed in Microsoft’s support documentation. TechCrunch frames the consolidation as an acknowledgment that Microsoft’s prior strategy was too complicated to make Copilot competitive with ChatGPT, Claude, and Gemini, following similar moves in the AI app space by Claude, OpenAI, and Google.

Filed from: TechCrunch .

OpenAI’s Computer History Records Clicks and Typing to Build ChatGPT Memories

OpenAI added an opt-in Computer History feature to the ChatGPT desktop app on macOS that records interaction events — clicks, typing, keyboard shortcuts, app switches, and context exposed through macOS accessibility — and periodically converts them into text summaries and local memory files that inform future chats. It is off by default and available to ChatGPT Pro, Business, and Enterprise users, but not in the EEA, Switzerland, or the UK; OpenAI says event files are stored locally unencrypted for up to 48 hours and are not used for training, while its documentation warns the feature increases prompt-injection risk from content in apps and websites.

Filed from: The Register , OpenAI documentation .

Claude Code Returns Blank Thinking Blocks, but Reasoning Still Costs You

Developers report that Anthropic’s API has been returning empty or truncated thinking-block summaries for Claude Opus 4.8 and Sonnet 5 even when summarized output is explicitly requested, with one GitHub issue documenting empty thinking blocks since July 16, 2026. Anthropic told The Register the matter is under investigation. Its documentation separately states that customers are billed for all thinking tokens generated even when collapsed or redacted, so a summary that fails to stream still counts toward output-token billing.

Filed from: The Register , GitHub issue .

Apple Trained Its Own AI Model for China With Help From Alibaba

Apple has developed a custom large language model for the China market with support from Alibaba, Reuters reported, citing three people familiar with the matter, in a departure from Apple’s previous practice of licensing domestic Chinese models. Reuters reported the move would make Apple the first US company approved to offer a proprietary AI model in China, with Apple Intelligence expected to reach China in the coming months after an iOS update. The report is based on unnamed sources and Apple did not respond to The Verge’s request for comment.

Filed from: The Verge , Reuters .

OpenAI Introduces Ultrafast, a Mode That Runs GPT-5.6 Sol at 14x Speed

OpenAI rolled out a new Ultrafast mode for ChatGPT that it says lets its flagship GPT-5.6 Sol model work at 14 times the speed of standard processing, delivering up to 750 output tokens per second. “Until now, getting real-time speed typically meant choosing a smaller or more specialized model,” the company said in a blog post. “Ultrafast points to progress in a new direction: more useful work per second.”

Filed from: TechCrunch .