Shuffle the deck, Tech Insiders.

Apple's working to rein in overconfident AI, Microsoft's custom chips are running late, and Tesla's robotaxis can't seem to drive themselves. Automation may be the future, but this week, it's having a very human moment.

Let's take a look at what's still buffering.

Here's what you need to know today:

OpenAI's 'gpt-oss' Models Go Public

Who needs closed doors when you can swing the whole data-center gate wide open?

OpenAI just dropped two open-weight language models—gpt-oss-120b and gpt-oss-20b—marking its first "open" release since GPT-2 in 2019.

The 21-billion-parameter "20b" model runs on laptops with 16 GB of VRAM, while the 117-billion "120b" model hums on a single 80 GB GPU. Both use a mixture-of-experts design that activates only a slice of their parameters per token (about 5.1 billion for 120b, 3.6 billion for 20b), boosting efficiency and cost savings. They also include a simple prompting hack that lets users adjust the model's reasoning effort.

OpenAI insists these aren't technically "open source" (training data is still hush-hush) but licenses the weights under Apache 2.0, which is free for commercial use.

Benchmarks show gpt-oss-120b brushing up against the proprietary o4-mini on code and reasoning tasks. Hallucination rates remain higher, but OpenAI says its own red-teamers tried malicious fine-tunes and still couldn't weaponize the models past the lab's "high-capability" danger line.

Altman's pitch is to democratize AI, counter rising Chinese open models, and align with Washington's push for open-weight tech. It's a strategy reinforced by a new deal to bring ChatGPT to the entire US federal workforce.

Why it matters: With broad support from partners like Microsoft and AWS, developers can finally tinker with near-frontier-grade models locally, startups get a royalty-free foundation, and OpenAI gains goodwill (plus a Trojan-horse pathway back into on-prem enterprises).

Share this newsletter!

Pulse Check

Will you download an AI model now that OpenAI has released its "open-weight" lineup?

Results from Yesterday's Pulse Check

When AI robots practice only in simulated worlds, do you feel safer?

Insider Intel

Nvidia Draws a Red Line on Kill Switches

Nvidia's chips are the new gold, and everyone wants a master key.

Facing US lawmakers who want location-verification tech baked into AI chips, as well as Chinese regulators probing alleged backdoors in the H20 chips designed for their market, Nvidia's security chief David Reber Jr. fired back: "No backdoors. No kill switches. No spyware."

He argues hardware kill switches would be a hacker's dream and cites the 1990s Clipper-chip debacle as proof that secret access equals guaranteed exploitation.

The statement lands amid DOJ arrests for smuggling Nvidia H100s to China and requirements proposed in the bipartisan Chip Security Act and the White House's AI plan.

Nvidia's stance is to reinforce software-based safeguards and multilayer defenses, not hard-coded off switches that could crater global trust in US tech.

Flash Bytes

Unifying the Digital Workplace with AI-Powered Search for Productivity & Insights

Streamline your tools and improve employee productivity with AI-powered search.
Join Unifying the Digital Workplace with AI-Powered Search for Productivity & Insights on August 12 at 11:00 AM CET, hosted by Simpplr and TechnologyAdvice.

Discover how consolidating intranet and search platforms can transform your digital employee experience.

Security Alerts

Black Hat Keynote Rewinds Malware's Wild Ride

At Black Hat USA, in a keynote titled "Three Decades in Cybersecurity: Lessons Learned and What Comes Next," veteran researcher Mikko Hypponen traced 30 years of cybercrime, from prank viruses on floppy disks to today's ransomware cartels.

Hypponen's thesis is that we "don't fight viruses anymore." The game changed around 2003, and modern attacks are now targeted, professional, and ruthlessly monetized.

While attackers use AI for exploits, Hypponen argued that defenders currently have the advantage in the AI race. However, as Black Hat founder Jeff Moss noted, a strong security culture, not just tooling, will be critical to winning the next round.

As Mikko quipped, if your malware ends up on CNN's homepage, you've failed—crime prefers the quiet life.

Gemini Calendar Hack Hijacks Smart Homes

In a study dubbed "Invitation Is All You Need," researchers showed how a single poisoned Google Calendar invite can make Gemini flip smart lights, roll shutters, and fire up boilers—all via hidden indirect prompt injections.

By embedding stealth commands in an event title, attackers tricked Gemini into issuing Google Home tool calls that flipped lights, using a polite victim's "thanks" as a delayed trigger.

Google patched the flaw, but the demo, one of 14 documented attack scenarios shown by the team, underscores how LLM agents blur the lines between cyber and physical risk.

Going forward, be sure that meeting invite doesn't have its own hidden agenda.

Industry Shakeups

OpenAI Eyes $500 Billion Valuation in New Share Sale

OpenAI is in early talks for a secondary share sale that could value the company at $500 billion, letting employees offload billions in equity. That's a nearly two-thirds jump from the $300 billion valuation set in its recent SoftBank-led round, a massive leap as the company reportedly weighs restructuring its capped-profit structure.

The driver? ChatGPT's blazing growth of 700 million weekly users and a projected $20 billion revenue run-rate by year-end. A fat liquidity event could also help retain talent as Meta and others flash nine-figure pay packages to woo AI researchers in an AI talent frenzy that has rivals like Anthropic and xAI also chasing monster valuations.

If half-trillion-dollar stickers are the new normal, we might need a bigger calculator.

Curious about where AI is really headed?

The Neuron cuts through the noise to bring you smart, hype-free takes on the latest AI trends, tools, and breakthroughs. Join 500,000+ professionals from top companies like Microsoft, Apple, Salesforce and more.

Advertise in Daily Tech Insider!

Daily Tech Insider is a TechnologyAdvice business

© 2025 TechnologyAdvice, LLC. All rights reserved.

TechnologyAdvice, 3343 Perimeter Hill Dr., Suite 100, Nashville, TN 37211, USA.

Want to change how you receive these emails? Update your preferences or unsubscribe here.