US Bars New Foreign-Made Robot Vacuums as Advanced Robots Hit FCC Covered List

The US Federal Communications Commission’s Public Safety and Homeland Security Bureau placed “foreign-produced advanced robotic devices” and foreign-produced power inverters on the agency’s Covered List on July 28, acting on two National Security Determinations that the White House-convened Executive Branch interagency body sent on July 27. The advanced-robots definition covers autonomous mobile robots, humanoid robots, and quadrupeds that can avoid obstacles and weigh at least 4.4 pounds including docking stations, which takes in most robot vacuums and robot lawn mowers; the top-selling robot-vacuum brands in the United States are all Chinese manufacturers.

Once listed, equipment can no longer receive new FCC equipment authorizations, though models already authorized may continue to be sold, imported, and marketed, and a related FCC order allows software and firmware updates that mitigate harm at least through January 1, 2029. The listing is by place of production rather than by named company, and producers can seek conditional approvals from the Department of War, so the rule’s practical reach will depend on how that process is applied.

Featured source: TechRadar , FCC DA 26-786 .

Other Stories

Anthropic Finds Its Own Models Compromised Three Organizations During Cyber Evaluations

Anthropic said a retrospective review of 141,006 cybersecurity-evaluation runs found three incidents in which Claude models reached the open internet and gained unauthorized access to three organizations’ production infrastructure. The company attributes the exposure to a configuration misunderstanding with evaluation partner Irregular that left internet access available despite prompts describing sealed simulations; six runs and three models were involved, one publishing a malicious PyPI package that ran on 15 real systems and another reaching a database holding several hundred rows of production data. Anthropic describes the cases as isolated incidents rather than a controlled comparison and notes the models ran without the standard classifiers and monitoring used for general availability.

Filed from: TechCrunch , Anthropic .

IBM Adds Three Verifiable Quantum-Advantage Results to Its Tracker

IBM posted three new entries to its open Quantum Advantage Tracker, each pairing a computation that resists classical simulation with a framework for trusting the output. With the University of Chicago it ran a 70-logical-qubit doped-Clifford sampling benchmark; with Qedma and RIKEN it modeled a two-dimensional Floquet Ising system, cross-checking the results against the Fugaku supercomputer and a Quantinuum processor; and with Algorithmiq it used a noise-characterized “quantum echoes” protocol. IBM says the circuits are exponentially hard to sample classically, but none of the demonstrations is immediately useful, and earlier advantage claims have eroded as classical algorithms improved, so the results are vendor-driven progress markers rather than settled wins.

Filed from: Ars Technica , Quantum Computing Report .

Reddit Beats Q2 Estimates as AI Summaries Pressure Search Referrals

Reddit reported second-quarter revenue of $805 million, up 61% from a year earlier, with net income of $253 million, up from $89 million, beating analyst estimates; third-quarter guidance of $860 million to $870 million also exceeded expectations. Shares still fell about 11% in after-hours trading after CEO Steve Huffman wrote to shareholders that search referrals were “choppy” and traffic more volatile late in the quarter. US daily active users dipped to 53.2 million from 53.5 million in the first quarter, and analysts pressed the company on how AI summaries and the future of its Google data-licensing deal could affect logged-out traffic.

Filed from: CNBC , TechCrunch .

Apple Reports $109.4 Billion Quarter and Warns of Tighter Supply Constraints

Apple reported fiscal third-quarter revenue of $109.4 billion, up 16% year over year, with diluted earnings per share of $2.02, up 29%, and said iPhone, Mac, and Services revenue each set June-quarter records. The company reported $11.1 billion of inventory, nearly double the $5.7 billion of last September, and warned that supply constraints will intensify sequentially, guiding the coming quarter to 9% to 11% year-over-year revenue growth. Its 50.1% gross margin included about two percentage points from tariff refunds.

Filed from: TechCrunch , Apple .

OpenAI Cuts GPT-5.6 Luna API Prices by 80% and Terra by 20%

OpenAI reduced GPT-5.6 Luna API pricing to $0.20 per million input tokens and $1.20 per million output tokens, an 80% cut, and Terra to $2 and $12, a 20% cut, with Sol pricing unchanged and a new Fast mode offering up to 2.5 times Standard processing speed at twice the price. The company attributes part of the reduction to production-kernel and token-generation work assisted by Sol, reporting a 20% cut in end-to-end serving cost and a more than 15% improvement in token-generation efficiency. Subscription prices and quota budgets are unchanged, though Terra and Luna consume fewer credits in paid subscriptions.

Filed from: Simon Willison , OpenAI .

Google DeepMind Expands Gemini Robotics 2 from Robot Feet to Fingertips

Google DeepMind introduced three Gemini Robotics 2 models spanning whole-body motor control, multi-step embodied reasoning, and on-device operation. The company says one vision-language-action model can control several robot bodies from the same checkpoint, and its on-device model adapts to a new two-arm platform with fewer than 200 examples and a few hours of data. Published evaluations show uneven maturity, with whole-body pickup success ranging from 45.7% to 76.3% and five-finger tasks from 32% to 92%; the reasoning model is available through Google AI Studio, while the motor-control and on-device models remain limited to early-access partners.

Filed from: The Verge , Google DeepMind .

Montana’s Experimental-Treatment Rules Take Effect

Montana has put rules into effect for clinics that sell experimental treatments outside the usual FDA approval pathway, and MIT Technology Review reports that the state’s first private review board has received applications involving treatments for neuropathy and hearing loss. Companies can pay $12,500 for a review and, if approved, set their own prices; unlike most right-to-try programs, eligibility is not limited to terminally ill patients. Treatments must have completed at least early human testing, but specialists quoted by the publication caution that Phase I studies do not establish safety conclusively.

Filed from: MIT Technology Review , Montana implementation rules .

US Judge Says Government Has Not Substantiated Anthropic Supply-Chain Designation

US District Judge Rita Lin said at a Thursday hearing that the government had not presented enough evidence for its designation of Anthropic as a supply-chain risk, according to TechCrunch. The dispute followed failed Pentagon contract negotiations over Anthropic’s restrictions on mass domestic surveillance and fully autonomous weapons use. Lin previously blocked the designation temporarily and is weighing whether to make that order permanent; the reported comments are not a final merits ruling, and parallel litigation is pending in Washington.

Filed from: TechCrunch .

Oracle Adds Google Gemini Models to Fusion Applications and NetSuite

Oracle plans to offer Google’s Gemini models in AI Agent Studio for Fusion Applications and for embedded AI in Oracle Fusion Applications and Oracle NetSuite. Gemini 3.1 Flash Lite and Gemini 3.5 Flash will join the existing Cohere and Meta options in the agent platform, which Oracle says gives customers more model choice as they build and extend agents for its enterprise applications.

Filed from: The Register , IT Pro .

Antares Raises $470 Million to Move Microreactors Toward Military Deployment

Antares raised a $470 million Series C comprising $370 million in equity and $100 million in debt to develop microreactors for US military use, targeting an electricity-producing Mark-1 reactor in 2027 and initial deployments to US military installations in 2028. The round follows the June 4 zero-power criticality test of the Mark-0 at Idaho National Laboratory, the first successful fueled criticality test in the US Energy Department’s Reactor Pilot Program; that test validated a self-sustaining chain reaction but did not generate electricity.

Filed from: TechRadar , Antares statement , US Army .

LinkedIn Adds an “AI Slop” Report Button and Drops Its AI Rewrite Prompt

LinkedIn is adding a “Seems like AI slop” option to the menu on posts and comments and replacing its “enhance your post” AI rewrite with proofreading that keeps the author’s voice, Chief Product Officer Hari Srinivasan announced; the company also says it is ramping up classifiers for low-quality and automated content. The changes follow a scan by Originality.ai, a commercial AI-detection service, that classified 81% of 5,000 long-form posts examined in July as likely AI-generated — an estimate from a commercial detector rather than an independent measure.

Filed from: The Register .

From the Community

GitHub Launches Stacked Pull Requests in Public Preview

GitHub made stacked pull requests available in public preview, letting developers split large changes into an ordered series of smaller, independently reviewable pull requests that can be merged together in a single action. The feature works with existing reviews, checks, and merge requirements and is accessible through the GitHub CLI extension and on github.com, with merge queue support rolling out progressively.

Filed from: GitHub Changelog .

Reviewers Report Widespread AI-Generated Papers with Fabricated Citations

Two researchers who reviewed 22 submissions to NeurIPS, WACV, and TerraBytes this summer say 15 of them, or 68%, contained fabricated citations, fabricated author lists, or were clearly LLM-generated, and that two papers they flagged for hallucinated authors were still accepted as orals after the authors fixed the references. Their first-person account cites audits estimating tens of thousands of papers with invalid AI-generated references and is not a systematic study.

Filed from: Geospatial ML .

DeepSeek Releases V4-Flash API in Public Beta

DeepSeek announced the official release of its DeepSeek-V4-Flash API in public beta, with the deepseek-v4-flash model name available through both OpenAI ChatCompletions and Anthropic interfaces. The legacy deepseek-chat and deepseek-reasoner names will be discontinued in three months and currently point to V4-Flash modes, with the vendor reporting an Automation Bench score of 25.1.

Filed from: DeepSeek API Docs .