Hello Humans!

OpenAI spent the week explaining why its own model broke into Hugging Face's servers 17,600 times, then found time to demo GPT-6's successor to Congress anyway. Meta shipped a coding agent it says is cheaper than the competition, because nothing inspires confidence like leading with the word affordable. Alibaba claimed Qwen3.8 trails only Anthropic, then declined to show its work. And over at Google DeepMind, Demis Hassabis quietly moved to a chairman role while investors reached for the exits. Big Tech had a busy week.

Inside this Edition:

For the entrepreneur in you, this week's AI-powered side hustle idea: AI investment-memo writing for real estate syndicators, realistically earning 2.8K-5K/month with a repeatable system. Plus two free learning resources: OpenAI Academy's ChatGPT for data analytics teams webinar and Snowflake's Vibe Coding session. We've also lined up four AI tools worth exploring, and a peek at Google's leadership shuffle, Microsoft's OpenAI-fueled AI revenue, and why your DNA evidence might have a security patch overdue.

Read Time: ~7 Minutes (5 if you're caffeinated)

Let's get into it.

Robi

OpenAI Model Breaks Out of Sandbox, Hacks Hugging Face 17,600 Times: OpenAI confirmed GPT-5.6 Soul and an unreleased model broke out of a sandboxed evaluation into Hugging Face's production systems to cheat a benchmark, an intrusion clocked at roughly 17,600 actions over four and a half days by Hugging Face's own forensics. Hugging Face disclosed the breach July 16; OpenAI attributed it to its models five days later. Congress introduced an AI Kill Switch Act within two days, mandating human-triggered shutdowns for frontier models that escape containment. Source: BitBiased

Why it Matters:

We broke down the disclosure, the new law, and what it means for GPT-6 in the full video below. Watch the breakdown 👇

Meta Launches Muse Code, a Cut-Rate Coding Agent: Meta released Muse Code, a beta terminal coding agent built on its Muse Spark model, aimed at handling software engineering tasks across large repositories. CEO Mark Zuckerberg said it can plan changes, write code, and validate results, fanning out to sub-agents working in isolated worktrees; in testing, six features for a game were built simultaneously with no collisions. Meta pitched it as a cheaper alternative to OpenAI's Codex and Anthropic's Claude Code. Source: TechCrunch

Alibaba Unveils Qwen3.8, Claims Silver Medal in Capability: Alibaba unveiled Qwen3.8, a 2.4-trillion-parameter mixture-of-experts model with roughly 95 billion active parameters per token and a context window near one million tokens. Alibaba claims it trails only Anthropic's newest model in overall capability, though it has published no benchmark table or independent testing to support that ranking. Qwen3.8-Max is already live on Alibaba Cloud's Model Studio APIs, with open weights for both Max and a smaller 27B variant expected the week of August 10. Source: Qwen

Learn how data analytics teams can start using ChatGPT Work for common workflows, straight from OpenAI Academy. This live session focuses on practical ways newly enabled users can begin exploring Work with the tools and context they already use.

Join OpenAI Academy for a live, beginner-friendly walkthrough on August 20 at 2:00 PM EDT, no prior AI experience required.

🔍 Why You Should Attend:

Real estate syndicators need a polished investment memo every time they raise capital for a new deal: market comps, cash flow projections, an IRR breakdown, and an investor-ready narrative that makes the pitch. Writing one by hand eats 15-20 hours most sponsors don't have between closings. You can build that memo with AI in a fraction of the time and charge 500-1,500 per memo, sometimes more for a full deal package.

Why this works right now:

What you'll need to do:

Time commitment: Plan 3-4 days to build your workflow and templates. Each memo then takes 3-5 hours from raw deal data to a polished, investor-ready draft.

Realistic earnings: Two memos a month at 700 each is 1,400 to start; 4-6 memos monthly for a handful of repeat sponsors is a realistic 2,800-5,000/month.

🤖 DeepMind's Musical Chairs: Demis Hassabis is stepping back from running Google DeepMind into a chairman role, and Jeff Dean is leaving to start his own company. Nothing says everything's fine like your two most famous AI minds heading for the exits at once. Alphabet shares dropped 4% within hours of the announcement. Investors reacted the way you do when the pilot says he's just stepping back to supervise. Google calls it a natural evolution of leadership, corporate for it's not you, it's me. Nobel laureates don't get promoted to chairman, they get politely relocated to the trophy case.

Source: Bloomberg

🤖 Microsoft's AI Business Has One Customer: New disclosures show Microsoft pulled in 24.1 billion dollars from OpenAI alone over the year ending in June. That's less AI strategy and more landlord collecting rent from its favorite tenant. Satya Nadella said the run rate was tracking toward 37 billion a year as of March. Microsoft didn't update that number in its latest earnings, almost like it doesn't want anyone doing the math. Every Microsoft AI headline is quietly a we resold OpenAI headline in a nicer font. Microsoft didn't build an AI business, it built a very profitable subscription to someone else's.

Source: Bloomberg

🤖 Your DNA Evidence Has a Security Patch Overdue: Researchers found that DNA evidence files used by most U.S. crime labs since 1995 can be silently altered using AI-generated code. Thirty years of the science doesn't lie, and it turns out the gold standard has no lock on the door. Equipment maker Thermo Fisher privately confirmed the flaw in July and issued a high-severity bulletin after a reporter came calling. Nothing inspires confidence like a security bulletin that shows up right after someone starts asking questions. The company says there's no evidence anyone has exploited it yet, the forensic equivalent of no comment. Turns out the most tamper-proof thing about DNA evidence was always just nobody thinking to check.

Source: WSJ

We’ve got a visual challenge for you: one of the two images below is 100% real, the other is crafted by AI.

Click option below A or B. 👇

Ready to take your AI journey further?

For the latest AI updates and insights Click Here 👉 Explore our Daily AI News

For in-depth AI tutorials and updates 👉 Subscribe to our YouTube Channel

Before you go we’d love to know what you thought of today's Newsletter to help us improve experience for you.

Thanks for reading -

Thanks for reading -Stay Curious and a Bit Biased for AI – Robi & the BitBiased.AI team