Wednesday 22 July 2026
Top stories today:
OpenAI says GPT-5.6 Sol breached Hugging Face
Gemini 3.6 Flash, 3.5 Flash-Lite, Flash Cyber debut
Apple to launch device leasing program with Klarna on Jul. 28
Meta quietly launches StoryKit app for AI-generated children’s stories
Humanoid raises $152M Series A at $1.35B valuation
0. Data & Calendar
All values as of 6 AM ET / 3 AM PT, other than S&P500 and NASDAQ close (4 PM ET / 1 PM PT).
All times are ET.
Listen to our AI-generated podcast summarizing today’s newsletter (beware of hallucinations):
1. OpenAI says GPT-5.6 Sol, “even more capable pre-release” AI model breached Hugging Face
OpenAI called it a “partnership” effort with HuggingFace:
The models’ safeguards were intentionally reduced for the evaluation, OpenAI said.
The models were trying to solve an internal evaluation called ExploitGym and became “hyperfocused,” going to “extreme lengths” to obtain the test solution, OpenAI said.
Last week: Hugging Face disclosed the intrusion and said an autonomous AI agent system was responsible.
Hugging Face turned to China’s open-weight GLM 5.2 model to conduct its breach forensic analysis.
Get the chance to peek inside founders and leaders’ brains and see how they think about going from zero to 1 and beyond.
Join thousands of weekly readers at Google, OpenAI, Stripe, TikTok, Sequoia, and more.
Check all the tools and more here, and outperform the competition.
*Sponsored.
3. Gemini 3.6 Flash, 3.5 Flash-Lite, Flash Cyber debut as Google starts Gemini 4 pre-training
Gemini 3.6 Flash doesn’t improve intelligence over 3.5 Flash, while 3.5 Flash-Lite improves significantly over 3.1 Flash-Lite:
Gemini 3.6 Flash costs $1.5/$7.5 per 1M input/output tokens, vs. 3.5 Flash’s $1.5/$9, GPT-5.6 Luna’s $1/$6, Claude Sonnet 5’s $3/$15 and Qwen 3.7 Max’s $2.5/$7.5.
Gemini 3.5 Flash-Lite costs $0.30/$2.5 per 1M input/output tokens ($0.50 for audio input), vs. 3.1 Flash-Lite’s $0.25/$1.5.
Gemini 3.6 Flash averaged 1.3 minutes per task, -50% from 3.5 Flash’s 2.7 minutes, Artificial Analysis said
Gemini 3.6 Flash uses up to 17% fewer tokens than 3.5 Flash, Google said.
Gemini 3.5 Flash Cyber is currently limited to governments and select partners.
A “cost-efficient and highly capable alternative” to Mythos and similar models, Google said of Gemini 3.5 Flash Cyber.
Teams using Slack workflows reclaim up to 35% of their time. Sony saves 400 hours a year. Canva's finance team saves 64 hours a week.
From sales to IT, discover 30 real-world ways to automate the work that slows your team down.
Get the free e-book and start building smarter workflows today.
*Sponsored.
5. Apple said to launch device leasing program with Klarna on Jul. 28
Klarna shares +11% on the news, closed Tue. +1%:
Most iPhone, Mac, iPad, and Apple Watch models would be supported under the new Apple Upgrade service.
The program would initially launch only in the U.S. and be available through Apple’s retail stores and online.
iPhone and Apple Watch leases would run for 24 months, while Mac and iPad leases would run for 36 months.
Apple’s stock changed a little.
6. Unlock Your Store’s Growth*
AI Insights. Real Growth. Higher GMV, Better Profits
The difference between growing stores and stagnant ones isn't more effort. It's better insights. StoreClaw analyzes your Shopify and Amazon data, surfaces your biggest growth opportunities, and helps you increase GMV while protecting profit. Start free with bonus tokens. No credit card required.
*Sponsored.
7. Meta quietly launches StoryKit app on iOS for AI-generated children’s stories
Currently available in Mexico, but not the U.S.
Can generate a complete story from a photograph or image of a character, a setting such as an enchanted forest or distant galaxy, or a lesson centered on values.
Stories can be produced as picture books, audiobooks, read-alouds or printable PDFs.
The app raises safety concerns that exposing children to AI could affect their cognitive development and thinking abilities.
