.png&w=3840&q=75)
AI News Today July 13 2026: Top 10 Stories
Two days after Apple sued OpenAI, the two most famous men in AI started brawling about it in public. And somehow that was not even the most important thing that happened this weekend. Google and Microsoft quietly teamed up against Anthropic and OpenAI, a new voice AI made talking to machines feel human for the first time, and the countdown to Gemini 3.5 Pro hit four days. I read all of it so you only need five minutes. Here are today's top 10 AI stories, in plain English.
1. Musk and Altman Turn the Apple Lawsuit Into a Public Brawl
Elon Musk and Sam Altman spent the weekend trading shots on X after Apple sued OpenAI on July 11 for trade secret theft. The lawsuit centers on a wild number: more than 400 former Apple employees now work at OpenAI, many from the teams that design Apple's chips and on-device AI. Musk, whose SpaceXAI competes with OpenAI and who has his own lawsuits running against Altman, amplified Apple's case and mocked OpenAI's hiring habits. Altman fired back. The tech internet grabbed popcorn.
Beyond the noise, the fight is getting bigger, not smaller. The Wall Street Journal reports Apple is preparing more countermeasures against OpenAI beyond the courtroom. And there is a delicious irony in the details: much of the chip technology at the center of the dispute came out of Apple's failed self-driving car project. The talent Apple could not keep busy is now the talent it is suing over.
Remember the stakes here. OpenAI wants to file for a $730 billion IPO as soon as September, and nothing spooks IPO investors like an unpredictable lawsuit from the most litigious hardware company on Earth. We covered the original filing in our July 12 roundup, and this weekend added the personality war on top.
My take: Musk jumping in is not random. Every day the news is about OpenAI's legal problems is a good day for his Grok models. In AI, even the feuds are strategy.
2. Google and Microsoft Team Up Against Anthropic and OpenAI
Google, Microsoft, Salesforce, Snowflake, and ServiceNow agreed to support a shared technical standard for connecting AI agents to business software, according to The Information, in a move aimed squarely at Anthropic and OpenAI. The battle is over something invisible but enormous: the plumbing that lets AI agents plug into company data and tools. Anthropic's standard, called MCP (Model Context Protocol), has quietly become the default over the past 18 months, and the giants do not love building on a competitor's foundation.
Why should you care about plumbing? Because standards wars decide who owns a technology era. The companies in this new alliance run the software where most of the world's business data already lives: Salesforce for customers, Snowflake for data, ServiceNow for workflows, plus the two biggest clouds. If they ship a common standard, every company deploying AI agents gets a Google-and-Microsoft-blessed alternative to Anthropic's.
Here is the twist that makes 2026 so strange: all of these companies, including OpenAI and Anthropic, are simultaneously members of a Linux Foundation group building shared open standards for AI agents. They are cooperating in the standards body and fighting in the market, at the same time, over the same technology.
My take: committee-designed standards move slowly, and MCP's head start is bigger than the headlines suggest. But when this much enterprise muscle lines up on one side, developers should at least hedge their bets.
3. OpenAI Ships GPT-Live, a Voice AI That Talks Like a Person
OpenAI released GPT-Live this week, a voice AI built on what engineers call a full-duplex architecture. Translation: it listens, thinks, and speaks at the same time, like a human does, instead of the walkie-talkie turn-taking that has made every voice assistant since Siri feel robotic. You can interrupt it mid-sentence and it adjusts. It also does real-time translation, searches the web while talking to you, and hands tasks off to other AI agents.
The turn-taking problem sounds small but it is the whole reason voice AI has felt fake for a decade. Humans overlap, interrupt, and react in milliseconds. A model that processes your voice while generating its own can hold a conversation at human rhythm, and early testers say the difference is immediately obvious. The first industry in line: call centers, where a natural-sounding AI that translates live changes the economics of the entire business.
The timing is pointed. Voice is where OpenAI's consumer lead is strongest, and Google (with Gemini built into Android and a big launch four days away) is its most dangerous rival there. Shipping GPT-Live now plants a flag on the one territory a Gemini price cut cannot capture.
My take: coding benchmarks are for developers, but voice is the interface for everyone else. The next AI war will be judged by ordinary people's ears, and that is a war OpenAI clearly intends to start ahead.
4. Anthropic's AI Is Now Guarding Critical Software in 15 Countries
Anthropic tripled its Project Glasswing this week, expanding from 50 partner organizations to 150 across 15 countries. Glasswing deploys Claude Mythos, Anthropic's restricted-access cybersecurity model, to find and fix vulnerabilities in software that societies genuinely depend on: utilities, hospitals, banks, and open-source projects too underfunded to audit their own code.
The backdrop is genuinely alarming. Intelligence agencies from the Five Eyes alliance warned in June that AI will transform cyberattacks within months, not years, and security firm Sysdig has already documented JADEPUFFER, the first ransomware operation run end-to-end by an AI. The uncomfortable truth: the same kind of model that can autonomously break into systems is the only tool fast enough to defend them. Glasswing is the defense side of that arms race.
The concentration question deserves an honest mention: 150 critical organizations now rely on one company's model and process for their most sensitive security work. That is a lot of trust in one vendor. It is also, given the alternative of machine-speed attacks against human-speed defenses, probably the right trade for now.
My take: this is the least flashy story of the week and arguably the most important. The AI security arms race is not coming, it is here, and most companies have not noticed yet.
5. Cloudflare Is Building a Cash Register for the AI-Powered Web
Cloudflare opened the waitlist for its Monetization Gateway, a system that lets websites charge AI agents for access, instantly and automatically. It runs on a standard called x402, which revives a dusty corner of the web's original design: HTTP status code 402, 'Payment Required,' reserved in the 1990s and never used. Now an AI agent hits a website, gets a machine-readable price, pays programmatically, and proceeds. No account, no checkout page, no human.
This matters because the web's old business model is dying in real time. Google Search went fully AI-generated on July 10, which means websites increasingly get read by machines instead of visited by people. No visits, no ads, no revenue. Machine payments are the leading candidate to replace that: instead of hoping an AI cites you, you charge it at the door. Cloudflare already sits in front of roughly a fifth of the web, making it the natural toll collector.
The open question is pricing. If access costs fractions of a cent, the AI-powered web stays abundant and everyone gets paid a little. If publishers price defensively high, we get a paywalled wasteland where agents can only afford the big sites. Nobody knows yet which way it tips.
My take: quietly, this might be the biggest business-model shift of the decade. The web is getting a machine-to-machine economy, and the companies that figure out agent pricing first will write the rules.
6. AI Can Now Edit a Whole Video From One Changed Image
A new workflow pairing OpenAI's GPT Image 2 with Runway's Aleph 2.0 lets editors change a single frame of a video (swap an outfit, relight the scene, replace an object) and have that change automatically carry through every other frame. What used to be weeks of frame-by-frame visual effects work becomes an image edit plus a button.
The technical wall this breaks is called temporal consistency. Editing one frame has been easy for years. Making 3,000 consecutive frames agree with each other is why VFX artists bill by the week, and why AI video tools stayed toys for professionals. Propagating one edited frame through a whole shot is the old Hollywood tracking pipeline rebuilt with AI doing the tedious part.
A grain of salt is required: demo videos always flatter these tools, and the hard cases (fast motion, things blocking other things, mirrors and reflections) are exactly what demos avoid showing. But the direction is unmistakable, and the commercial market for editing existing footage is arguably bigger than the market for generating new video, because studios and brands sit on mountains of footage they want to change, not replace.
My take: the VFX industry just watched its billable hours get compressed the way translation and stock photography did. The artists who master these tools will do the work of ten. The rest have a hard conversation coming.
7. Anthropic Will Pay You to Learn AI, No Degree Required
Anthropic announced Claude Corps, a paid 12-month fellowship that trains people as AI professionals by embedding them inside nonprofit organizations. The eligibility rules are the headline: you need to be 18 or older, have less than two years of work experience, and have US work authorization. That is it. No degree required.
Read those requirements again, because they target a very specific person: the entry-level worker that the AI economy is currently squeezing hardest. Companies have spent two years cutting exactly the junior roles that used to be the first rung of a career, and this week's labor data (story 9) shows how much anger that is building. Claude Corps is one of the first industry programs that actually aims at that rung: paid, year-long, hands-on, and open to people without credentials.
The nonprofits win too. Charities and community organizations cannot pay AI-engineer salaries, so they have been locked out of the productivity boom entirely. A fellow embedded for a year builds them real systems, not a slide deck about AI readiness.
My take: I am usually cynical about corporate fellowships, but the no-degree requirement is genuinely radical for this industry. If the cohort sizes are real and not token, this is a model worth copying. If you know someone early in their career, send them this.
8. Goldman Sachs Tells Wall Street Which Chinese AI Models to Use
Goldman Sachs published research recommending specific Chinese AI models to its clients, per CNBC. Let that sink in: the most establishment bank in America is now advising companies on which Chinese AI to deploy. The models have earned it on merit. DeepSeek V4, Kimi K2.6, GLM-5, and Qwen3.5 hold four of the top five spots in open-weight AI globally, and Alibaba shipped its newest Qwen3.6-Max-Preview this month.
The math driving the recommendation is brutal and simple. Chinese open-weight models deliver roughly 80 to 90 percent of frontier capability at a fraction of the price (DeepSeek's output costs around $0.44 per million tokens versus $30 for GPT-5.6 Sol, roughly a 70x difference). For the boring, high-volume work that makes up most business AI usage (summarizing, classifying, extracting), paying frontier prices is increasingly hard to justify to a CFO.
The geopolitical whiplash is real, though. This lands weeks after OpenAI, Anthropic, and Google teamed up to block Chinese labs from copying their models, and while Washington keeps tightening chip export rules. American finance is simultaneously funding the defense against Chinese AI and recommending it to clients.
My take: money finds the value, always. The interesting question is not whether enterprises will use Chinese models (they already do), but how long Washington pretends otherwise.
9. Most Workers Now Want a Share of AI Profits
A majority of surveyed workers support creating a wealth-redistribution fund financed by AI profits, per CNBC, as tech layoffs continue accelerating through 2026. The same weekend, Fortune reported a wave of early retirements among tech veterans who would rather leave than rebuild their careers around AI, while the Guardian found software engineers responding the third way: aggressive retraining and, increasingly, organizing.
Three reactions, one underlying fact: the AI economy's gains and pains are landing on different people. Productivity is up and company profits are strong, but entry-level hiring in exposed fields has been evaporating, and workers can see both lines on the chart. When a majority of workers back redistribution, that is not a fringe idea anymore, it is polling that eventually becomes a campaign platform.
The institutions are starting to respond. The Federal Reserve stood up its first AI task force this week (controversially co-led by venture capitalist Marc Andreessen), the European Central Bank warned AI could destabilize inflation, and programs like Claude Corps (story 7) target the broken bottom rung directly. Whether any of it moves fast enough is the open question.
My take: the industry spent two years saying AI creates more jobs than it destroys. Workers spent those two years watching the job postings. The polling gap between those two experiences is now the most important number in AI politics.
10. Gemini 3.5 Pro Is Four Days Away, and Google Cannot Miss
Google's Gemini 3.5 Pro launches July 17 per leaked plans, now just four days out, and the pressure on this launch is unlike anything Google has shipped before. The specs remain mouthwatering: a 2-million-token context window (double anyone else, roughly 30 novels in a single prompt), a new Deep Think reasoning mode on the $250 per month Ultra plan, and API pricing around $1.25 per million input tokens, a quarter of what OpenAI charges for GPT-5.6 Sol.
But the model is six weeks late, and the week it lands in has been merciless: GPT-5.6 launched July 9, Grok 4.5 on July 8, and GPT-Live voice just this week. Google also spent June losing two of its biggest stars, with Gemini co-lead Noam Shazeer leaving for OpenAI and Nobel laureate John Jumper for Anthropic. A great launch erases all of those headlines. A mediocre one confirms them.
Three things have to go right: it must beat GPT-5.6 Sol on at least one benchmark that matters, the giant context window has to actually work at full length (long-document recall that falls apart halfway would be worse than not shipping it), and it has to arrive on the 17th. On the hopeful side, Google Search already runs entirely on Gemini 3.5 Flash, which proves the infrastructure can handle planetary scale.
My take: my prediction, held loosely: Gemini wins on price and context, splits the benchmarks, and the real verdict comes two weeks later when developers with huge documents either migrate or do not. Thursday will be fun either way.
Frequently Asked Questions
Q: What did Musk and Altman fight about?
Elon Musk and Sam Altman traded public barbs on X over the July 11-12 weekend after Apple sued OpenAI for trade secret theft tied to hiring more than 400 former Apple employees. Musk amplified the lawsuit and criticized OpenAI's hiring; Altman fired back. The two also have their own long-running legal disputes.
Q: What is GPT-Live?
GPT-Live is OpenAI's new voice AI that listens, thinks, and speaks simultaneously (a design called full-duplex), so conversations flow at human rhythm and you can interrupt it naturally. It also handles real-time translation, live web search during conversation, and task handoffs to other AI agents.
Q: When does Gemini 3.5 Pro come out?
Leaked plans point to July 17, 2026, four days after this post. Expected specs include a 2-million-token context window, a Deep Think reasoning mode on the $250 per month Ultra tier, and API pricing near $1.25 per million input tokens. Google has not officially confirmed the date.
Q: Are Google and Microsoft actually working together on AI?
On one specific front, yes. The Information reports Google, Microsoft, Salesforce, Snowflake, and ServiceNow agreed to back a shared standard for connecting AI agents to business software, countering Anthropic's widely adopted MCP standard. The same companies still compete fiercely everywhere else.
Q: What is Claude Corps and who can apply?
Claude Corps is Anthropic's paid 12-month fellowship that places early-career people inside nonprofits to build AI systems. Eligibility: 18 or older, less than two years of work experience, and US work authorization. No degree is required.
Q: Can AI really edit an entire video from one image?
Yes, within limits. A workflow pairing OpenAI's GPT Image 2 with Runway's Aleph 2.0 lets an editor modify one reference frame and propagate the change across the whole video automatically. Hard cases like fast motion, occlusion, and reflections remain the weak spots.
Q: Which Chinese AI models did Goldman Sachs recommend?
Goldman's analysis, reported by CNBC, points clients to the leading Chinese open-weight models. The current top tier is DeepSeek V4, Kimi K2.6, GLM-5, and Qwen3.5, which hold four of the top five open-weight positions globally at a fraction of frontier pricing.
Q: What is Anthropic's Project Glasswing?
Project Glasswing deploys Anthropic's restricted Claude Mythos cybersecurity model to find and fix vulnerabilities in critical software like utilities, hospitals, and open-source projects. It expanded this week from 50 partner organizations to 150 across 15 countries.
Recommended Reads
• Top 10 AI News: July 12 2026 Daily Roundup
• Top 10 AI News: July 10 2026 Daily Roundup
• Top 10 AI News: July 9 2026 Daily Roundup
• Top 10 AI News: July 8 2026 Daily Roundup
Weekends like this are why keeping up with AI feels like a second job. Five focused minutes a day beats a panicked Monday catch-up, every time.
References
• The Information: Google and Microsoft team up on agent protocol
• Tom's Hardware: Agentic AI Foundation under the Linux Foundation
• Medium: AI news week of July 6 to July 12, 2026
• Third Run Time: Daily AI headlines for July 12, 2026
• TechCrunch: OpenAI launches the GPT-5.6 family
• Tech Brew: OpenAI, Anthropic, Google unite against distillation
.png&w=3840&q=75)
.png&w=3840&q=75)