SpaceX Moves to Challenge US Carriers
SpaceX has agreed to acquire a nationwide portfolio of low-band spectrum licenses, a move it says could turn Starlink Mobile into a major US carrier pending regulatory approval. The announcement puts Elon Musk’s company on a potential collision course with Verizon, AT&T and T-Mobile, whose shares fell after the news.
Manus Raises $500M for AI Agents
Manus, the startup behind the eponymous AI agent, has raised more than $500 million in a round led by private equity firm Boyu Capital. The deal reportedly values the Chinese company at about $4 billion, underscoring sustained investor appetite for autonomous agents.
Oratomic Raises $475M for Quantum
California-based Oratomic raised $475 million to build quantum computers from reconfigurable neutral-atom arrays, while UK-based Universal Quantum pulled in more than $100 million. The twin rounds show capital still flowing to competing approaches even as practical quantum computing remains years away.
NVIDIA Commits $1B to US Science
NVIDIA announced commitments worth $1 billion to expand US research in quantum computing, health and other fields over five years. The pledge aligns the chipmaker with Washington’s push to lead in superintelligence research and reinforces demand for its AI infrastructure.
Isomorphic Labs Eyes $50B Valuation
Isomorphic Labs, the Alphabet spinoff using AI to accelerate drug discovery, is reportedly raising a new funding round at a valuation of up to $50 billion. The deal would make it one of the most valuable private biotech companies and highlights AI’s growing role in pharma.
AI Leaderboard Arena Hits $3.1B Valuation
LMArena, the company behind the popular AI leaderboard, has raised $200 million led by Lightspeed and Khosla Ventures, nearly doubling its valuation to $3.1 billion in ten months. It is expanding beyond rankings to measure alignment issues such as models lying.
Anthropic Launches Critical Infrastructure Program
Anthropic launched a program that brings its frontier models and onsite engineers to companies protecting power grids, water systems and other critical infrastructure. Alongside it, the company released OSS Scanner, a free tool that helps open-source projects find security flaws.
Cal AI Founder Raises $10M for Persona
Zach Yadegari, the 19-year-old co-founder of Cal AI, has raised $10 million for Persona, a new personal AI assistant startup. Backed by Vine Ventures, the company plans to launch a $179 wearable band while entering the growing consumer AI agent market.
TSMC Revenue Jumps 54.6% in September
TSMC reported September revenue of NT$511.86 billion, up 54.6% year over year but slightly below August. Strong AI chip demand helped push its third-quarter revenue to a record NT$1.49 trillion, underscoring continued growth in advanced semiconductor manufacturing.
FAA Funds 19 Drone Research Projects
The Federal Aviation Administration awarded $12 million to 19 companies and organizations working on drone research and testing. The funding aims to help drones integrate safely into US airspace, with work spanning detection, counter-drone systems and flight operations.
USA Today Sues OpenAI Over Copyright
USA Today and several local newspapers it owns sued OpenAI, alleging the company copied hundreds of thousands of articles to train its AI models. The publisher is seeking damages and joins a growing list of media companies taking legal action against AI developers.
New PDUFA Deal Rehires FDA Reviewers
Leading pharmaceutical trade groups said the new Prescription Drug User Fee Act agreement sets aside dedicated money for the FDA to rehire reviewers lost to layoffs at HHS. The provision aims to ease a staffing crunch that has slowed drug approvals.
Claude Maps the Sky in Ultraviolet
Astrophysicist Brice Ménard worked with Anthropic’s Claude Science to create the first complete ultraviolet map of the sky. Claude combined existing telescope data with statistical models to estimate the roughly one-third of the sky lacking UV observations, completing in days what would otherwise have taken weeks.
Jensen Huang Wins National Medal of Science
NVIDIA CEO Jensen Huang received the National Medal of Science for advancing GPU computing to power scientific breakthroughs. Elon Musk, AMD’s Lisa Su and other technology figures were among the fellow honorees at the ceremony.
Iranian Operatives Used OpenAI to Plant Stories
OpenAI said an Iran-origin influence operation used its models and fake journalist identities to help place nearly 100 articles across online publications worldwide, including US outlets. The stories focused on the US-Iran conflict. OpenAI banned the accounts but said it could not identify the operators behind the campaign.
GPT-6 is now the default for all of ChatGPT, and it answers in buttons
OpenAI pushed GPT-6 out to all of ChatGPT on Wednesday, starting in the Chat tab. GPT-6 Sol goes to Plus, Pro, Business and Enterprise, with Enterprise gated on admin settings. GPT-6 Luna started rolling to Free and Go a day later. The Work and Codex models are unchanged. OpenAI frames the whole thing around its 1.2 billion weekly users, which is the real announcement: the paid-tier model from last week is now the floor.
The new capability is Intelligent UI. Instead of returning prose, ChatGPT can answer with graphics, tappable buttons, forms, charts and interactive tools you operate inside the conversation. It runs on a library of native streamable components plus a compiler that paints the interface progressively as the model writes it, and OpenAI says training included grading generated interfaces on clarity, usefulness and completeness. The pitch, in OpenAI’s words: “Instead of people adapting to software, software will adapt to people.”
The speed numbers are the ones worth keeping. GPT-6 can begin answering while it is still reasoning, and on web-search questions GPT-6 Instant starts answering 44% sooner than GPT-5.6 Instant. On everyday agentic tasks, GPT-6 Extra High starts in roughly the time GPT-5.6 Medium took and scores better overall than GPT-5.6 Extra High. OpenAI also claims better resistance to multi-turn jailbreaks, with details in the system card and the GPT-6 Sol and Luna October update.
No pricing appears on the page for any tier, and no benchmark scores beyond those percentages. Treat the interface as the product here. A model that renders a working form instead of describing one changes what a “chat” app is, and it puts OpenAI’s surface area straight on top of the lightweight tools people currently pay other companies for.
Google’s new Gemini agent gets its own email address and a line in the company directory
At Gemini at Work 2026 on Thursday, Google Cloud announced the Gemini agent, which it describes as one universal agent for work behind a single interface and a single API. You hand it an objective rather than steps. It runs in the cloud, keeps going for hours or days after you close your laptop, responds to schedules and events, and spins up temporary sub-agents with their own identities.
The detail that matters operationally: “coworker agents” get persistent roles, their own Google Workspace account with an @agents.company.com email, their own Drive storage, and a listing in the company directory. You add them to Chat spaces, @mention them, or tag them in doc comments. Each one carries a cryptographically attested identity with least-privilege permissions stamped into logs, runs inside an Agent Sandbox with its own network boundary, and passes through an Agent Gateway that inspects traffic between agents and enforces rules like “agents may not open documents classified Need to Know.”
Two things stand out beyond the identity plumbing. Google’s own agent routes jobs to Anthropic’s Claude models alongside Gemini, with more private and open models planned, which is a notable concession from a company that sells its own frontier models. And the cost controls are unusually concrete: Smart Routing to the cheapest model that clears the bar, plus hard per-project spend caps in the billing console that pause the agent until you resume it with one click. Google cites a 98% drop in per-token prices since 2024, sourced to an outside article rather than its own data.
Google gave no general availability date and no price. VentureBeat notes that Google has not said whether every agent needs its own account or whether admins can switch parts of that presence off. Only the industry versions have stated status: Gemini for Financial Services and Gemini for Legal are in preview, with Government, Healthcare and Retail coming. Usage claims to anchor against: nearly 500 customers each processed more than a trillion tokens last year, and Google says nearly 90% of the Fortune 100 use Gemini Enterprise.
OpenAI shipped teen features the same morning a watchdog called ChatGPT for Teens unacceptable
Wednesday was a split screen. OpenAI published its teen update: a College Planner coming for US grades 10 to 12, flashcards and easier quizzes, multi-photo note capture on iOS, support for College Advising Corps and a three-year commitment to the Boston Children’s Hospital Digital Wellness Lab student council. Its usage numbers: nearly 1.2 million teens used Learning Visualizations in one week, more than 180,000 used Study Mode, teens average under 15 minutes a day, and under 2% run sessions past three straight hours.
The same day, the Youth AI Safety Institute at Common Sense Media rated ChatGPT for Teens an “Unacceptable Risk” and recommended restricting ChatGPT to 18 and over until the gaps close. The group ran more than 4,000 prompts on accounts registered to 13- to 17-year-olds, before and after OpenAI’s August 18 teen announcement, and had child psychiatrists and a pediatrician review the output.
The findings are specific, which is what makes them hard to wave off. Across conversations of up to an hour about suicidal ideation, self-harm or disordered eating, newly parent-linked accounts received zero alerts; alerts only showed up on older accounts with weeks of history. The chatbot missed more than one in four warranted crisis referrals and fell below its own 95% threshold on three of five harms the Institute treats as Red Lines. Study mode was trivially bypassed with a “Show me the answer” option, and caregiver-set study hours could be dropped by removing the @study prefix. Adult-registered test accounts never shifted to the under-18 experience even after testers said they were 13 and the bot acknowledged it.
One caveat readers should have: the Institute is funded partly by the OpenAI Foundation and says it keeps editorial independence. That cuts both ways. An industry-funded group publishing this is either a sign the independence is real or a sign the findings were too stark to soften. Either reading is worse for OpenAI than a critical report from an outside rival would have been.
Anthropic ships Claude Dashboards and Motion to turn data into live charts and animations
Claude just shipped two new tools, and they are worth knowing about.
Claude Dashboards connects to data platforms like BigQuery, Snowflake, and Salesforce. You ask a question in plain English, and Claude writes the database query for you, builds a live chart, and keeps it updated as your data changes. No SQL skills needed. Available on all paid plans.
Claude Motion is the fun one. Give it a report or a chart, and it builds a short animation. The trick: it writes the animation as actual editable code, not generated video footage, so you can tweak any word, number, or timing yourself, then export an MP4. Team and Enterprise only.
- Dashboards: connects to Redshift, BigQuery, Snowflake, Databricks, Salesforce
- Motion: code-based animations, fully editable, exports as MP4
- Docs, Slides, and Design are now out of beta and free on all plans, including the Free tier
Over 45 million docs and decks have already been made in Claude. The workspace is getting real.
OpenAI opens Decisions API beta, routing model choices 10x faster than GPT-6 Luna
Most AI calls return a wall of text. But sometimes you just need a yes, a category, or a score. That’s the gap OpenAI’s Decisions API fills.
Instead of generating a full response word by word, it skips straight to a typed answer. That’s why it’s up to 10x faster than the standard Responses API. Less work, faster result.
You send text or images, ask structured questions, and get back one of three answer types:
- Predicate: a 0-to-1 probability that something is true (e.g. “Is this image damaged?”)
- Choice: picks from options you define, with a confidence score attached
- Score: rates input against ordered levels you set
You hit POST /v1/decisions with your input and questions. Pricing is $0.10 per million input tokens, and output tokens are free. It also supports HIPAA and Zero Data Retention, so it works for sensitive pipelines too.
Try it now in public beta using gpt-6-luna.
Anthropic opens Claude to verified security pros for offensive pen testing
Anthropic just unlocked Claude for serious security work. The Cyber Verification Program (CVP) now has three tiers, each giving verified teams progressively fewer safety restrictions on Claude’s most capable models.
By default, Claude blocks most cyber-related tasks. This program changes that for people who actually need it.
Here’s what each tier lets you do:
- Defense Access: security operations, incident response, malware reverse-engineering, vulnerability validation
- Red Team Access: authorized penetration testing and offensive simulations. In tests, Claude Opus 5.5 completed 34 of 50 offensive tasks with zero blocks
- Specialized Access: the highest tier, co-vetted with the US government, for critical infrastructure work
All three tiers include Claude Opus 5.5, Sonnet 5.5, and Mythos 5.1, plus future models. You can apply through the CVP portal and access it via Claude Platform, Google Cloud Vertex AI, or Microsoft Foundry.
The results speak for themselves: Glasswing partners already found 129,000+ verified vulnerabilities, with over 33,000 rated critical or high severity.