Secure the cockpit, Tech Insiders. Microsoft is rolling its own AI engines while PACs file flight plans to steer AI policy, and breaches try to ground the fleet.
Buckle in—we're clearing turbulent airspace. |
|
|
Here's what you need to know today: |
|
|
Microsoft's MAI Models Break Cover |
Redmond finally bakes its own brain, no OpenAI required.
After years of riding shotgun with OpenAI, Microsoft unveiled two in-house power plants: MAI-1-preview (text) and MAI-Voice-1 (speech).
The text model, trained on a lean 15,000 NVIDIA H100s, hit 13th on the LMArena leaderboard and will soon integrate into Copilot chats. MAI-Voice-1, meanwhile, cracks out 60-second audio clips in under a second on one GPU, powering Copilot's daily briefings and podcasts (and is testable in Copilot Labs).
AI chief Mustafa Suleyman refers to the duo as Microsoft's first homegrown models and teases an upgrade using its now-operational NVIDIA GB200 cluster. The strategy: blend in-house AIs with partners to keep Copilot nimble and trim its OpenAI dependency. |
Image Source: Gemini (Veo 3) |
Early benchmarks position MAI-1 below top models from OpenAI and Anthropic, but it's "punching way above its weight" in terms of efficiency, especially since xAI's Grok was trained on over 100,000 H100s.
Under the hood, its mixture-of-experts architecture fires select parameters to cut GPU burn, taming cloud costs as AI demand explodes.
Microsoft wasn't alone last week. The launch sparked a news blitz that saw OpenAI unveil a rival voice model, xAI introduce a hyper-efficient coding AI, and Anthropic encourage users to opt in for data training. Still, Microsoft insists its OpenAI marriage is intact… just a little more "open."
Why it matters: Microsoft's $13 billion OpenAI fling bought an early AI head start, but the crown requires owning the engine room. If MAI can deliver GPT-class quality at a low cost, Copilot could scale to billions of users without exceeding Azure's budget or its GPU resources. |
|
|
Would you let an AI voice assistant read your daily news instead of a human anchor? |
|
|
Results from Friday's Pulse Check |
Should chatbots alert someone when users mention plans of self-harm? |
|
|
UK Workers Sweat Over AI Pink Slips |
With the UK government pushing for rapid AI deployment, union chiefs are advocating for a "worker first" AI blueprint: attach strings to public tech funding, share productivity gains, and give employees equal say in deployment decisions. The TUC also wants beefier social safety nets and skills programs to cushion inevitable disruption.
Critics warn that without guardrails, AI could exacerbate inequality as shareholders reap the benefits and entry-level job opportunities disappear.
Across the pond, Goldman Sachs predicts any AI job-loss wave will be short-lived. However, CEOs from OpenAI to Meta and Anthropic continue to forecast entire classes of roles disappearing by the 2030s. Translation: reskill now, argue later.
Somewhere, a robot HR rep is drafting a "don't panic" memo, probably in perfect AP style. |
|
|
TransUnion Salesforce Hack Hits 4.4M |
Credit-reporting giant TransUnion said a July 28 breach of its Salesforce instance via a third-party app exposed names, birth dates, and Social Security numbers for 4.4 million US consumers. Core credit data was (allegedly) untouched, but leaked support tickets reveal plenty of PII. |
Passwordstate Bug Opens Emergency Backdoor |
Click Studios patched a high-severity authentication bypass (disclosed Aug. 28, CVE pending), letting attackers craft a single URL to waltz past Passwordstate's Emergency Access gate and straight into admin controls. Build 9972 fixes the flaw and hardens the browser extension against clickjacking.
With 29,000 organizations and 370,000 security pros trusting Passwordstate for crown-jewel credentials, delay is not an option. This marks the fourth authentication bypass vulnerability for the software since 2022, a track record that also includes the infamous 2021 supply-chain hack that injected malware via updates.
If you absolutely can't patch yet, the official workaround is to lock down the Emergency Access page by IP address. The good news? A successful exploit at least triggers an email alert, so you'll know the fox is in the henhouse. It seems Passwordstate's most consistent feature is finding new ways to test its own authentication... in production.
|
Silicon Valley PACs Gear Up for 2026 |
Crypto déjà vu: AI heavyweights just formed Leading the Future, a super PAC network seeded with over $100 million from VC firm Andreessen Horowitz, OpenAI president Greg Brockman, and pals to back pro-innovation candidates in 2026.
Not to be outdone, Meta is launching its own California-focused PAC with a name only a PR team could love, pledging tens of millions to nudge Sacramento toward lighter AI rules before the next governor's race. |
The playbook mirrors Fairshake's crypto blitz, which helped shape D.C. sentiment in 2024: flooding primaries with ad cash, elevating AI-friendly lawmakers, and countering "AI doomer" bills that could splinter state regulations.
Run by a bipartisan team—Fairshake spokesman Josh Vlasto, a former Schumer aide, and Targeted Victory CEO Zac Moffatt, a Romney campaign veteran—the group's early targets are New York, California, Illinois, and Ohio.
Expect glossy ads touting job creation and national-security stakes, while pro-regulation "doomers" form their own PACs and warn of regulatory capture in hoodie clothing. Silicon Valley once told Washington to "move fast and break things." Now it's filing FEC paperwork to make sure nobody fixes them afterward. |
|
|
|
Writer at TechnologyAdvice |
Justin Meyers is an investigative writer and editor who draws on over a decade of meticulous hands-on research to deliver the full, trustworthy story behind consumer and enterprise tech, including cybersecurity. |
|
|
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 215, Nashville, TN 37211, USA. |
|
|
|
