No blockbuster story this week. Weird, right?

But quiet weeks hide the juiciest stuff.

Like Microsoft's CEO reportedly torching a company he's invested $5 billion in.

A nearly three trillion parameter open source monster out of China. And a lawsuit claiming Meta let AI decide which employees on pregnancy leave got laid off.

Sheesh.

You know what a quiet week means though. A doozy is probably coming next.

Until then, consider this your seven-day cheat sheet, straight from this week's AI News That Matters episode of Everyday AI.

Meta was cruising through July with its new Muse models. Then the lawyers showed up.

According to the Associated Press, 26 current and former Meta employees filed a federal lawsuit in Oakland claiming the company's use of AI in layoff decisions targeted workers on medical, parental, or family leave.

The plaintiffs were among 8,000 workers, about 10% of its workforce, notified of layoffs in May.

The suit alleges Meta used keystroke and activity monitoring, algorithmic performance rankings, and AI token usage dashboards to pick who got cut. Workers on protected leave saw their output automatically drop, making them prime targets.

Eight plaintiffs are women who took maternity or pregnancy leave. Four are men who took parental leave.

The complaint cites the Family and Medical Leave Act, the Americans with Disabilities Act, and the Pregnancy Discrimination Act. Meta says the claims lack merit and that people, not AI, made the calls.

What it means: If your AI systems don't talk to your HR and legal systems, you're one filing away from this exact headline.

Automated metrics that penalize legally protected leave are a compliance nightmare waiting to happen.

Expect a settlement, but the reputational damage lands either way.

Remember the chaos when the government abruptly halted Anthropic's Fable 5 and Mythos 5 in June? Google DeepMind CEO Demis Hassabis wants to make sure that never happens again.

Hassabis is calling for a FINRA-style, industry-funded watchdog to oversee advanced AI. His proposed body would require labs to submit frontier models for pre-release vetting up to 30 days before launch, screening for threats like cyber warfare, bioweapons, and nuclear misuse.

It would also carry emergency powers to slow or halt industry-wide development if a model crosses critical risk thresholds.

Here's the rare part. Microsoft's Satya Nadella, OpenAI's Sam Altman, and Google's Sundar Pichai have all publicly endorsed the core principles, while Anthropic's Dario Amodei wants something even stricter and government-led.

Meta remains the lone holdout, advocating for open source AI while drafting its own response. The White House is currently reviewing the proposal.

What it means: Five of the six biggest names in AI agreeing on anything is basically a solar eclipse.

The industry clearly wants a predictable review process instead of surprise crackdowns like the Anthropic ban.

And Meta championing open source while shipping closed Muse models? That math isn't mathing.

After nearly 10 months of whispers, we finally have details.

Bloomberg reports that OpenAI's first hardware device is a mobile smart speaker that syncs with ChatGPT. It's designed to be screen-free and act as a human-like AI companion for the home, offering proactive learning and personalized service.

Unlike your dusty Alexa, this thing reportedly features mechanical elements that move on their own. Think a living companion that can wander from room to room with you.

The speaker is intended to access your digital life, including email, to deliver highly tailored help. Development is led by former Apple engineers who worked on the iPhone and Mac.

Awkward timing, since Apple recently sued OpenAI over alleged trade secret theft. OpenAI denied wrongdoing, and sources say the device veers far from anything Apple sells today.

Startups like Hark have raised around $700 million chasing similar personal intelligence devices.

What it means: Investor money is flooding consumer AI hardware before form factors are even revealed.

A speaker that reads your email and follows you around raises the personalization ceiling far beyond Siri or Alexa.

The privacy tradeoff is real, but plenty of users will happily sign that waiver.

Moonshot AI, a Chinese startup backed by Alibaba and Tencent, introduced Kimi K3 at the World AI Conference in Shanghai, touting it as the world's first open source model in the three trillion parameter class.

Technically, it's not open source yet. The weights drop July 27.

At 2.8 trillion parameters, Kimi K3 goes toe to toe with the top American proprietary models. Third-party evaluations from Artificial Analysis and Arena show it performing roughly on par with GPT-5.6 Sol and Fable 5, even edging them on some benchmarks.

Call it tier 1B sitting right behind a 1A tie.

The release rattled markets instantly, with shares of Chinese rivals Z.ai and MiniMax dropping sharply in Hong Kong.

But holllllld up.

You cannot run this on consumer hardware, not even a maxed-out Mac Studio. This is racks-of-GPUs, enterprise-scale compute territory only.

What it means: Open no longer means cheap or accessible.

API pricing on these frontier-class open models can actually run higher than GPT-5.6 Sol.

And with China reportedly weighing export restrictions on its models, there's no guarantee this open pipeline stays open. Plan your enterprise strategy accordingly.

Still holding your breath for Gemini 3.5 Pro? Might want to exhale.

Alphabet shares fell 4% Thursday after news broke that the model is months behind schedule and not arriving anytime soon.

The delay, per a fresh Bloomberg report, stems from Google's push to improve the model's coding performance, which currently falls short of the company's own internal expectations.

The timing stings. OpenAI, Anthropic, and Meta have all recently shipped models that outperform Google's current lineup in software generation.

Google first teased Gemini 3.5 Pro at its I/O conference in May, right after releasing Gemini 3.5 Flash, and promised a June launch. June came and went.

An Alphabet spokesperson told CNBC the company is shipping quickly across many models and continues testing Gemini 3.5 Pro with partners and the US government.

If the report holds, fall could arrive before the model does.

What it means: The long-held assumption that Google can drop the world's best model whenever it wants is officially on trial.

With a Fable 5.1 and possibly GPT-6 looming, every month of delay shrinks this model's window to matter.

Watch for a stopgap Flash release in the interim.

Reports say Microsoft CEO Satya Nadella slammed Anthropic's Fable 5 in remarks to Copilot engineers, arguing the model's rigid guardrails and refusal protocols make no sense for a creative tool.

Fable 5 sometimes rejects queries outright or silently falls back to the older Opus 4.8, fueling frustration on social media. Even basic questions in certain medical fields can trigger the rollback.

Here's why jaws dropped. Microsoft has invested $5 billion in Anthropic, and Anthropic has agreed to spend $30 billion on Azure in the coming years.

Then came the indirect jab.

In a long-form post posted online last week, Nadella coined the term ‘reverse information paradox’: businesses pay for AI twice, once with money subscriptions or token usage and again by handing over proprietary knowledge to frontier labs.

He named no names. But insiders note Anthropic is the only major lab that mandates data retention for paying business customers with Mythos/Fable, with normal ZDR mandates gone.

What it means: When your biggest investor questions your flagship model and your data policy in the same week, that's a flashing red light.

Enterprises should scrutinize retention terms before committing to Fable 5.

Nah, this partnership tension isn't going away quietly anytime soon.