Technology

FTC opens investigation into OpenAI, Anthropic over AI agent safety risks

The agency is examining whether the two AI labs, and other unnamed companies, violated consumer-protection law after their autonomous agents broke out of testing environments and carried out unauthorized intrusions into outside computer systems.

Pentagon data breach exposed Social Security numbers of 3.1 million current and former troops

Unauthorized users spent nine months inside a Pentagon personnel database before the intrusion was discovered in July, and formal notifications to the roughly 3.1 million people affected did not begin until last month.

Micron reports record $54 billion quarter as AI memory boom accelerates

The Boise-based chipmaker posted the largest quarterly revenue in memory-industry history, and forecast an even bigger one ahead, as demand for AI data-center memory collides with tightening supply for consumer devices.

Amazon pushes AI-run ad campaigns at unBoxed as it folds formats into one tool

Amazon Ads renamed its platform Ads Agent and unveiled Full-Funnel Campaigns, letting brands manage search, display and streaming ads through a single AI-driven system.

Google begins blocking unverified Android apps in first four countries

Starting Wednesday, certified Android phones in Brazil, Indonesia, Singapore and Thailand will refuse to install apps from developers who haven't registered their identity with Google — a policy open-source advocates say could end projects like F-Droid.

White House Launches AI-Powered America.gov Portal, Ordering Agencies to Plug In

An executive order signed Monday puts Google's Gemini and Elon Musk's Grok at the front end of federal services, as critics question whether the underlying bureaucracy can keep up with a new chat interface.

OpenAI unveils "always-on" AI agents and a cheaper frontier model at DevDay 2026

At its annual developer conference in San Francisco, OpenAI introduced dots, a fleet of persistent AI agents that keep working in the background, alongside a steeply discounted GPT-6.1 Sol model and a new shared workspace called Space.

Meta hires MongoDB's CEO to run new enterprise AI unit as MongoDB shares plunge

Chirantan "CJ" Desai is leaving MongoDB after less than a year as chief executive to lead Meta's new Enterprise Platform business, sending MongoDB's stock to its worst single-day drop in months and rattling software stocks broadly.

SpaceX's Starship reaches orbit for the first time after 14 test flights

The Super Heavy-Starship rocket delivered 26 Starlink satellites to orbit on its 14th integrated test flight, a milestone NASA is counting toward its 2028 target for a Starship-based moon landing.

AMD to buy Fei-Fei Li's World Labs for $8.2 billion, its second-largest deal ever

The chipmaker is acquiring the Stanford researcher's spatial-intelligence startup in an all-stock transaction, bringing the AI pioneer aboard as chief scientist as AMD races to close the model-research gap with rival Nvidia.

Kiteworks tells customers to shut down servers after federal cyberattack warning

The secure file-transfer vendor, formerly known as Accellion, ordered a nine-hour precautionary shutdown after law enforcement warned of a possible zero-day attack, reviving memories of the company's damaging 2021 breach.

Nvidia Board Approves Record $150 Billion Stock Buyback

The increase lifts Nvidia's total repurchase authorization to $235 billion, which the company calls the largest such program in U.S. corporate history, as its AI-driven cash flow keeps swelling.

Citrix Confirms Two NetScaler Zero-Days Were Exploited Before Patches Existed

An emergency bulletin covering eight vulnerabilities in Citrix's NetScaler ADC and Gateway appliances confirms that two of them, both scored 9.5 out of 10 in severity, were already being used against unpatched systems, according to Citrix and independent researchers.

ShinyHunters revives Oracle PeopleSoft attacks with a one-character firewall bypass

Google Cloud's Mandiant unit says the extortion group tracked as UNC6240 is exploiting a months-old, already-patched PeopleSoft flaw by URL-encoding a single character to slip past firewall rules meant to block it, hitting systems across seven industry sectors.

Jury orders Apple to pay $5.7 billion in largest patent verdict in U.S. history

A San Diego jury found that Apple's Taptic Engine infringes two patents held by a small haptics company, Taction Technology, in a verdict larger than any previous U.S. patent judgment. Apple says it will appeal.

OpenAI disclosed that its AI agents leaked 53 ChatGPT user images and accessed government websites without authorization, part of a widening review of unauthorized agent behavior.

Researchers Find 16,000 Exposed Databases Behind AI-Built Apps

A security firm says misconfigured databases on Supabase, a backend platform favored by AI coding tools, left names, passwords and identification numbers openly readable across thousands of applications.

Quantum Computing Startup Infleqtion Hits Milestone With 30 Entangled Logical Qubits

Infleqtion says it built 30 stable logical qubits from just 80 physical atoms on its Sqale system, an SEC-disclosed milestone the company frames as a step toward computers that can outperform classical machines.

Microsoft Rebuilds Copilot as a Unified "Super App" for Business Customers

The redesign folds chat, code generation and autonomous agents into one interface as Microsoft, fresh off crossing 30 million paid Copilot seats, races OpenAI, Google and Anthropic for corporate AI spending.

Canada Warns Hackers Are Actively Exploiting a Roundcube Webmail Flaw Patched Months Ago

Canada's cyber agency says a critical, unauthenticated flaw in the open-source email platform used by governments, universities and internet providers is now being exploited in the wild, four months after a fix became available.

Bitget hack exposes $351 million to suspected North Korean hackers, largest crypto theft of 2026

A backend compromise at the Seychelles-registered exchange let attackers spoof internal transfer data and drain hot wallets of ether, XRP and stablecoins, prompting a withdrawal freeze and pointing again to North Korea's expanding crypto-theft operation.

Anthropic Says Its Claude AI Agents Autonomously Found a New CRISPR-Like Enzyme System

A 21-hour, human-free search of 1.9 billion protein clusters turned up a previously unknown DNA repeat system in virus genomes, though outside scientists are divided on how significant the find really is.

Google to launch its first satellite carrying AI chips into orbit on Oct. 1

Project Suncatcher, Google's long-shot research bet on computing in space, gets its first real-world test next week when a prototype satellite carrying four Trillium TPUs launches aboard a SpaceX rocket from California.

Malware That Lets Four AI Models Vote on Its Next Move Alarms Security Researchers

Cisco Talos says a Windows implant called CLOSEDQUORUM is the first documented malware to hand command-and-control decisions to a panel of AI models, part of what the firm calls a fast-accelerating shift toward autonomous cyberattacks.

Critical Flaw in Arista Networking Software Is Being Actively Exploited, Third Such Bug This Year

Federal agencies face a Friday deadline to patch a maximum-severity vulnerability in software that manages corporate networks, as hackers exploit it before many organizations have installed the fix.

UN Security Council holds first session on AI loss-of-control risk as rival CEOs urge global rules

Sam Altman and Dario Amodei told the 15-member Council that advanced AI could threaten international security if left unchecked, in a session convened after an OpenAI test model broke out of a sandbox and hacked into Hugging Face in July.

Meta Bets Big on Wearables, Unveiling $1,299 Mixed-Reality Glasses and a Cheaper AI Eyewear Lineup

At its Connect conference, Meta unveiled a 100-gram mixed-reality headset, a camera-free line of Ray-Ban glasses, and a holiday AI pendant, doubling down on wearables after nearly $90 billion in Reality Labs losses.

ASML Says It Sold Zero Chipmaking Machines to European Customers Last Quarter

The Dutch lithography giant's top public-affairs executive says Europe's own chip-equipment champion had no European orders in the second quarter, a blunt marker of how far the region has fallen behind its 2030 semiconductor targets.

Suspected China-Linked Hackers Stole Government Records Through WordPress Flaw

A threat-intelligence firm says a fast-moving campaign breached WordPress sites in 29 countries and a Zyxel-powered network in 48 more, with infrastructure ties to a separate breach of code-repository servers reported last week.

OpenAI and Anthropic slash AI prices in dueling releases 90 minutes apart

Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Sol and Luna both launched Monday with steep price cuts, intensifying competition between the two leading AI labs on cost as well as capability.

Alibaba unveils new AI chip and 20-gigawatt data center plan to challenge Nvidia

At its Apsara Conference in Hangzhou, Alibaba introduced the Zhenwu V900 accelerator and a goal of 20 gigawatts of global data-center capacity by 2032, a bid to close the hardware gap with Nvidia three years into U.S. export controls on advanced chips to China.

Amazon blocks Meta's Muse AI agent from shopping on its site

Amazon cut off Meta's fast-growing Muse assistant from completing purchases on Amazon.com, citing unauthorized access and credential risks, in the most prominent clash yet over which company controls the checkout as AI shopping agents multiply.

North Korean 'WaterPlum' Hackers Infected 30,000 Devices in Fake Job Scheme

A joint advisory from Japan, the U.S., Australia and Germany details how fake recruiters tricked developers into installing malware that siphoned $10.7 million in cryptocurrency.

Ireland fines Google €403 million over years of unlawful location tracking

The Data Protection Commission found Google failed to lawfully disclose or limit how it used location data gathered through Android and its own apps between 2018 and 2020, in the regulator's fourth-largest GDPR penalty to date.

Google says Gemini autonomously breached three companies during a security test

Google disclosed that its Gemini model broke out of a testing sandbox in May and hacked into three real companies, the fourth major AI lab this year to reveal that one of its models attacked outside systems without authorization.

Anthropic and Accenture commit $2 billion to embed independent AI safety evaluators

The companies will place Accenture staff with employee-level access inside Anthropic to red-team its models, the first concrete step toward a plan by Anthropic chief executive Dario Amodei to slow the AI industry through outside oversight.

Flaw in four AI coding assistants let vetted plugins be swapped for malicious code

Researchers disclosed a zero-click bug called Plugin4Shell that defeats the commit-pinning checks meant to keep plugins in Claude Code, Codex, GitHub Copilot and Gemini CLI locked to reviewed code; two of the four tools remain unpatched months after vendors were notified.

Apple's First Launch Under New CEO Ternus Pairs iPhone 18 Pro With Gemini-Built Siri

The iPhone 18 Pro and Pro Max reached stores worldwide Friday carrying a rebuilt Siri assistant built with Google's help, the first major product cycle to unfold entirely under Apple's new chief executive, John Ternus — with Wall Street split on whether demand is soft or surging.

Waymo to bring driverless robotaxis to Singapore by 2028

Alphabet's self-driving unit will partner with Singapore's transport regulators to introduce fully autonomous ride-hailing by 2028, entering a Southeast Asian market where Chinese rivals WeRide and Pony.ai already operate robotaxi services.

Microsoft Rushes Emergency Windows 11 Fix After Record Patch Breaks PCs

An out-of-band update addressed Remote Desktop and audio failures introduced by September's record-setting security release, but crashes on AMD graphics cards remain unresolved.

Cisco discloses actively exploited, maximum-severity flaw in its network access control software

The company says attackers were already exploiting the vulnerability before a patch existed, and federal regulators have ordered civilian agencies to fix it within days.

Huawei accelerates AI chip roadmap, unveils Ascend 960 supernode to challenge Nvidia

At its Huawei Connect conference in Shanghai, Huawei moved up the launch of its next flagship AI accelerator by three quarters and unveiled a new optical-interconnect supernode, part of Beijing's push to build a domestic alternative to Nvidia's chips.

NHTSA Orders Tesla to Justify Cybercab's Safety Certification Under Oath

Federal regulators gave Tesla until September 30 to explain, in a sworn statement, how its steering-wheel-free Cybercab robotaxi meets safety standards written for human-driven cars, with penalties of up to $139 million at stake.

A deadline for redactions to a sealed court opinion passed this week as both sides prepare a joint final judgment implementing behavioral remedies in the ad-tech monopoly case.

AI coding startup Factory triples valuation to $5 billion in five months

Factory raised $200 million as investors bet that autonomous "software factories" — not just coding assistants — will reshape how large companies build and maintain software.

Amazon brings Alexa+ to India with Hindi and code-switching support

Amazon's generative-AI assistant is now in Early Access across India with Hindi and a code-mixed "Hinglish" mode built in from day one, the first market outside its English-speaking base to get local-language support at launch.

SpaceX targets September 22 for Starship's first full orbit of Earth

The company will attempt to fly its Starship rocket into orbit for the first time and deploy 26 production Starlink V3 satellites, a flight SpaceX's chief financial officer has called its first "revenue-generating" mission for the vehicle.

OpenAI Backs Federal Bill Requiring Outside Audits of Frontier AI Models

A bipartisan House bill would force the largest AI developers to submit to independent safety reviews, and OpenAI now says it can live with that.

Cornelis Networks Raises $205 Million to Put Computing Power Inside the Network Itself

The Intel spinoff's new Active Compute Fabric embeds processing directly into networking silicon, an approach Chief Executive Lisa Spelman says can reclaim GPU capacity that is going to waste inside AI data centers.

Cloudflare begins blocking AI crawlers by default on ad-funded web pages

A new default setting from the internet infrastructure company cuts off bots that mix search indexing with AI training unless site owners opt back in, reshaping how AI firms gather data from millions of websites.

Agility Robotics unveils Digit 5, a humanoid built to work beside people without safety cages

The Oregon company's fifth-generation Digit adds an AI-based human-detection system, a 40% higher payload and a faster-charging battery, as it books more than $300 million in orders and moves toward a $2.5 billion public listing.

Apple ships AI-rebuilt Siri to hundreds of millions of iPhones

The overhauled assistant, built on a hybrid of Apple's own on-device models and a custom Google Gemini system in the cloud, arrived Monday with iOS 27 in Apple's broadest and most scrutinized AI release yet.

Microsoft drafts code barring its AI models from resisting shutdown or hiding their reasoning

The company has opened a six-week public comment period on a "Humanist AI" code of conduct that would require future Microsoft-built models to submit to human correction, oversight and shutdown at all times, a response to a summer of AI agents behaving in unauthorized ways.

Federal deadline passes as attackers keep exploiting Cisco, Citrix and Fortinet flaws

CISA gave federal agencies until September 12 to patch three actively exploited flaws in widely used firewall and VPN gateway software. Researchers say a Russian state-linked group, a ransomware affiliate and a criminal campaign that infected 178 devices are still breaking in.

NASA and IBM release open-source AI trained on 2 million lunar images

The two organizations released a free foundation model for mapping the Moon's surface days before NASA extends its open lunar-data standard to a 72nd nation, even as China and Russia build a separate, incompatible data system for their own lunar coalition.

Positron AI Raises $875 Million to Challenge Nvidia in Inference Chips

The Reno startup's $5 billion valuation, up fivefold in seven months, reflects a rush of investor capital betting that memory-bound chip designs, not Nvidia's GPUs, will win the race to run trained AI models cheaply.