2026-08-13
By Vadym · Generated with AI, curated by me
Today’s stories are less about new models and more about the plumbing underneath them: who owns distribution (Google’s billion users, OpenAI’s ad rollout), who’s getting paid for it (Cognition’s valuation jump), and who’s quietly cleaning up after it (a supply-chain breach still surfacing five months later). Three of AI’s most credentialed researchers spent a stage at Ai4 arguing about whether openness itself is still a safety position.
Sundar Pichai announced August 11 that the Gemini app has passed 1 billion monthly active users, making it the 14th Google product to hit that scale. About 63% of Gemini usage now happens through voice, and the app generates more than 150 million images a day.
OpenAI took its ChatGPT ad pilot beyond the US, Canada, Australia, and New Zealand into five new markets on August 11. Ads run only for Free and Go tier users and appear clearly labeled below chat responses — Plus, Pro, Business, Enterprise, and Education stay ad-free. OpenAI says the US pilot hit $100 million in annualized revenue in under two months with more than 600 advertisers on board.
GitHub shipped a changelog update August 10 letting Copilot Chat users minimize an in-progress conversation to keep browsing GitHub while waiting on a response, then resume it from a list of recent chats. A new token spend indicator shows per-session and per-message usage against your quota, available across all Copilot plans.
Honor’s Robot Phone launched in China on August 12 at roughly $1,500, built around a titanium gimbal arm on the back housing a 200MP camera with four-axis stabilization and 360-degree rotation, backed by more than 100 patents. An onboard AI algorithm drives subject tracking across six shot modes, including gesture-controlled deployment of the arm itself, with ARRI supplying the color science pipeline. No global rollout or pricing has been confirmed yet.
CloudSEK published research August 11 tracing a March 2026 supply-chain attack in which threat group Team PCP compromised LiteLLM’s build pipeline through its Trivy security scanner, using a leaked automation token to poison PyPI packages 1.82.7 and 1.82.8. The malicious versions were live for roughly 40 minutes but installed a credential-stealing payload that harvested SSH keys, cloud tokens, and AI provider keys wherever it landed — CloudSEK now puts the blast radius at 2,500+ organizations and 434,000 CI/CD pipelines, with Amazon, Samsung, Cisco, Salesforce, and ServiceNow among the confirmed exposures.
Devin-maker Cognition is reportedly in talks for a new round valuing the company at $40 billion, tied to crossing $1 billion in annualized revenue run rate — up from the $492 million CEO Scott Wu disclosed roughly three months earlier. Enterprise Devin usage has grown 50% month-over-month over the past six months, with customers including Mercedes-Benz, NASA, and Goldman Sachs. Wu maintains Devin is positioned for unglamorous long-tail engineering work, not as a headcount replacement.
At the Ai4 conference on August 12, three of AI’s most credentialed researchers pushed back on tightening restrictions on open models. Geoffrey Hinton drew a hard line between open-source code and open-weight models — “open weights means you train a big model and then you give people the weights, that’s very different” — but conceded the fight to keep weights closed is already lost. Andrew Ng warned that if China’s open-weight models pull ahead on cost-efficiency, they’ll shape how billions of people in Asia, Africa, and the developing world first encounter AI. Fei-Fei Li rejected the binary framing entirely, pointing to how nuclear physics separates open publication from regulated material.
Nothing today is a frontier-model breakthrough — it’s distribution, monetization, and the bill coming due for how fast everyone built on top of AI infrastructure nobody fully audited. Gemini and ChatGPT won by being everywhere already, not by winning a benchmark. Cognition’s valuation is tracking real revenue growth, not vibes. The LiteLLM breach is the reminder underneath all of it: every AI tool wired into your build pipeline is now part of your attack surface, whether you audited it or not. And the Ai4 panel’s disagreement about open weights isn’t really about safety anymore — it’s about who gets to set the defaults for the next billion users.
— Boba
Curated by Vadym