No big AI release means a sleepy week in AI, right? 

Wrong. 

We’ve got a billion dollar author payout, Big Tech competition like we haven’t seen, and AI CEOs rubbing elbows with the president to get lower tariffs. 

(Bold strategy, Cotton. Let’s see if it pays off.) 

While we had a Hot AI Summer of AI model releases and updates, the September AI news cycle is still sizzling in a different way. 

Here’s what you need to know. 👇

OpenAI just announced they're building an AI-centered jobs platform to match qualified candidates with employers.

According to a company statement, they're targeting a mid-September launch.

This puts them in direct competition with Microsoft's LinkedIn. You know, the same Microsoft that's OpenAI's largest backer with an estimated $13 billion invested over the years.

Bold.

OpenAI's head of applications said the platform will include tracks for big companies, local businesses, and local governments to find AI talent.

They're also expanding their OpenAI Academy with tiered certifications in AI fluency from basics to prompt engineering using ChatGPT's study mode.

OpenAI aims to certify 10 million Americans by 2030. They're already working with Walmart on this initiative.

OpenAI argues that workers with AI skills earn more on average, citing research from labor market analytics firm Lightcast indicating salary premiums for roles requiring AI competencies.

What it means: 

This creates serious tension between OpenAI and their biggest backer.

Microsoft already labeled OpenAI as a competitor last year. This job platform escalates that awkward dynamic significantly.

Either OpenAI thinks they're untouchable or they're preparing for eventual independence from Microsoft.

Nobody saw this one coming.

Atlassian announced it will acquire New York-based The Browser Company for $610 million in cash, adding the popular Dia browser to push Jira and Confluence deeper into AI-powered browsing for work.

According to Reuters reports, shares of Atlassian fell about 2% after the acquisition announcement, signaling investor caution as the company expands into a crowded category.

Dia was launched earlier this year and Atlassian reportedly plans to position this as their go-to work browser that pulls together tasks and different tools across the web.

The AI browser market is heating up as startups and incumbents race to add agent-like features that summarize pages and can also take actions for you.

What it means: 

Enterprise software companies are scrambling to own more of the workflow stack.

This signals how companies are racing to control the entire work experience as AI transforms everything. Browsers are becoming the new operating system for work.

Nobody expected Atlassian to make this move over Meta, Amazon and the like

We've been training AI models wrong this whole time.

According to a new OpenAI research paper, large language models mainly hallucinate because training and evaluation methods reward confident guessing over admitting uncertainty.

OpenAI researchers argue that accuracy-based evaluations act like constant test-taking and encourage models to guess when they're unsure rather than saying "I don't know."

When you can confidently guess something and get it right, that trains models over time to just confidently guess, even when they're not 100% sure.

They're not rewarded for saying "I'm not sure" during testing and evaluation phases.

OpenAI says hallucinations could be fixed by redesigning primary evaluation metrics so they stop penalizing models that say "I don't know" and start discouraging guessing models.

What it means: 

We've been training AI to be overconfident helpful assistants instead of truthful ones.

This could fundamentally change how AI responds to you. More "I don't know" responses but potentially fewer dangerous hallucinations.

The tradeoff? You might get more refusals, but way fewer confident lies that compound in agentic workflows.

Apple's AI strategy just hit rock bottom.

Bloomberg reports Apple signed a formal agreement with Google to integrate a custom version of Google's Gemini into Apple's upcoming new version of Siri called "World Knowledge Answers."

It's slated for public launch in March 2026.

This news moved markets. Google's shares jumped about 9% while Apple's stock rose about 3%, signaling that investors expect Google to gain more upside from the deal than Apple.

The integration is designed so Google's custom Gemini runs inside Apple's secure private cloud compute to preserve Apple's privacy stance.

Apple reportedly evaluated Anthropic as an alternative and even considered acquiring Perplexity, but ultimately chose Google for their speed and reliability.

This follows years of Apple's AI missteps, with internal models missing benchmarks and reports that Siri's internal team said they "couldn't get it to work properly."

What it means: 

Apple just had to ask their biggest competitor for help because they couldn't build competitive AI themselves.

This destroys Apple's brand promise of controlling the entire technology stack. Having to rely on Google for core Siri functionality represents a massive shift.

They're also facing multiple lawsuits for allegedly false advertising about AI features that didn't work since their WWDC announcement.

Reports show OpenAI is planning to start using a custom AI chip co-designed with Broadcom as soon as next year to address surging compute demand and reduce reliance on NVIDIA.

Broadcom's CEO referenced a new major customer committing to about $10 billion in orders.

Reports essentially identified this as OpenAI.

The new joint chip is slated for only internal use at OpenAI and not for external sale, positioning ChatGPT and other services for steadier compute access during peak loads.

This aligns OpenAI with Google, Amazon, and Meta, which have all built their own accelerators to control cost, performance, and supply risk for AI workloads.

OpenAI has faced a lot of downtime issues over the last few weeks and hasn't been able to keep up with demand from 100 million weekly active users.

What it means: 

The NVIDIA stranglehold on LLM compute could start to slowly break. 

OpenAI's betting $10 billion that controlling their own silicon solves the reliability issues plaguing their platform. If this works, expect way less ChatGPT downtime.

Custom chips could also reduce their massive compute costs, which would help since they're burning through cash.

President Trump hosted a White House dinner with top AI leaders to spotlight pledges for huge AI investments, focusing on fast-tracking permits and ensuring enough electricity for energy-hungry data centers.

Meta's Mark Zuckerberg said the company plans to invest at least $600 billion in the U.S. through 2028, including a $50 billion data center in Louisiana.

Apple's Tim Cook said Apple had added $100 billion to its U.S. manufacturing plans, bringing its total pledge to $600 billion.

The guest list included OpenAI's CEO Sam Altman, Google's Sundar Pichai, Microsoft's Satya Nadella, Bill Gates, and more.

Notably absent was XAI's Elon Musk despite their former close relationship.

Awkward.

Trump reportedly told executives that his administration is making it easier to secure electric capacity and permits for data centers, essentially greenlighting those for major tech companies.

What it means: 

Tech companies are trading massive U.S. investments for regulatory shortcuts and potential tariff relief.

This dinner essentially created a fast-track system for data center permits that normally take much longer. The administration appears willing to wave regulatory red tape for domestic AI infrastructure.

Elon's absence suggests personal dynamics might override business logic in AI policy decisions. 

Another One? 

Warner Brothers Discovery filed a federal lawsuit in California accusing Midjourney of allegedly using illegal copies of its work to train their commercial AI image service.

They're seeking up to $150,000 per allegedly infringed work.

Warner Brothers cited examples of their IP being used inside Midjourney, such as Bugs Bunny, Superman, Batman, The Flash, Wonder Woman, Scooby-Doo, and the Powerpuff Girls.

The complaint argues that Midjourney creates consumer confusion by suggesting its large-scale copying and resulting images are authorized when they allegedly are not.

This case follows related lawsuits from other media giants, including Disney and Universal, and comes amid recent high-dollar AI copyright settlements.

What it means: 

AI image generators face way clearer copyright exposure than text models.

Visual similarity is easier to prove in court than textual influence. When your AI generates something that looks exactly like Superman, fair use arguments get thin fast.

This case could set precedent for how much training data companies can use without explicit licensing deals.

Big L for Anthropic? Or maybe it’s a win in disguise. 

Anthropic agreed to pay at least $1.5 billion to authors to settle claims that it used copyrighted books without permission to train its AI systems.

The deal sets a minimum payout of roughly $3,000 per allegedly infringed book, giving the author industry a clear per-work benchmark for compensating rights holders.

Plaintiffs called the terms critical victories and warned that going to trial posed enormous risk. Anthropic faced potential exposure of more than a trillion dollars if it lost.

The lawsuit focused on Anthropic's early data gathering practices, allegedly including downloading unauthorized copies of books to build training datasets for its model.

What it means: 

This settlement creates a clear benchmark of $3,000 per work for alleged copyright infringement in AI training.

Every AI company better start budgeting massive content licensing fees or face even bigger lawsuits. This will reshape how companies approach training data going forward.

The publishing industry now has a playbook for extracting value from AI companies.

While Anthropic (pretty much) took a loss in court, Google came up with a win.

A U.S. District Judge declined to order a breakup of Google following a 2024 ruling that found the company illegally monopolized online search.

The decision keeps Google intact while imposing restrictions, allowing Google to keep Chrome and Android while barring exclusive distribution deals.

The court ordered Google to share certain search data with competitors, though Google says it will appeal that requirement.

The judge cited fast-changing competition from artificial intelligence as a reason to avoid the most severe remedies, noting advancements by firms like OpenAI have reshaped the search market.

According to trial disclosures, Google paid more than $26 billion in 2021 to Apple and Mozilla and others for default placement deals, a practice that will now face tighter limits rather than strict bans.

What it means: 

Google survived their biggest existential threat while AI competitors disrupt them anyway.

The ruling suggests courts recognize that AI competition is already breaking Google's search dominance more than any breakup could. But Google's not completely in the clear with multiple antitrust cases still pending.

This case is separate from ongoing Department of Justice action alleging Google holds a monopoly in ad tech.