Unitree Robotics Soars 629% in Record Shanghai IPO
Chinese robotics company Unitree Robotics delivered the year's most spectacular market debut on Tuesday, with shares surging 629% on Shanghai's STAR Market — the largest first-day gain for a tech listing on the exchange this year. The humanoid robot maker briefly touched a $66 billion valuation, making it the first publicly traded humanoid robot company on mainland China.
The IPO raised 6.1 billion yuan ($904 million) at a share price of 150.80 yuan, and its retail tranche was oversubscribed more than 8,000 times, setting a STAR Market record for technology startups. Unitree is already profitable, with 1.7 billion yuan in 2025 revenue and over 18,000 robots delivered — a rarity among robotics startups that typically burn cash for years before turning a profit.
The frenzy underscores investor conviction that humanoid robots are on the cusp of commercial viability, fueled by rapid advances in AI-powered motor control and computer vision. (Bloomberg, Forbes)
Anthropic's Claude Outperforms Human Experts at Protein Design
Anthropic revealed on Monday that its Claude models (Mythos Preview and Opus 4.8) successfully designed protein binders for 14 out of 15 targets — a critical first step in drug development that traditionally takes human specialists weeks or months per target.
The results are striking: between 22% and 35% of Claude's individual designs bound successfully, compared to the 10–15% success rate typical in professional protein design campaigns. Some of Claude's strongest designs bound several times more tightly than the best previously published results for those targets.
The research signals a potential paradigm shift in how early-stage drug discovery is conducted. If an AI model can match or exceed expert biochemists at designing candidate molecules, it could dramatically compress the timeline and cost of bringing new therapies to clinical trials. Anthropic framed the work as part of its broader Claude Science initiative to demonstrate real-world scientific capabilities. (Anthropic)
OpenAI Launches Teen-Safe ChatGPT With Parental Controls
OpenAI rolled out a dedicated ChatGPT for Teens experience on Monday, aimed at users aged 13 to 17, featuring a comprehensive suite of safety guardrails and parental oversight tools. The move comes amid mounting scrutiny over how young people interact with AI chatbots.
The teen version blocks conversations around self-harm, eating disorders, violence, and sexually explicit content. It introduces regular break reminders that encourage users to step away, and a Study Mode that steers teens toward working through homework rather than simply copy-pasting answers.
Parents can set "Quiet Hours" when ChatGPT is unavailable and receive safety notifications — reviewed by trained personnel — when high-risk interactions are detected. Notably, parents do not get blanket access to their teen's conversations, a deliberate privacy design choice. The rollout uses age-prediction technology to auto-route minors into the protected mode. (TechCrunch, OpenAI)
Stripe Closes $7B+ Deal to Acquire OpenRouter
Payments giant Stripe finalized its acquisition of OpenRouter for more than $7 billion — a staggering 5.4x markup from OpenRouter's $1.3 billion Series B valuation just three months prior. The deal, first reported by Bloomberg, positions Stripe at the intersection of AI infrastructure and financial services.
OpenRouter operates as an AI gateway that routes requests across 400+ AI models from OpenAI, Anthropic, Google, Meta, and DeepSeek, serving roughly 8 million developers. It lets customers pick the right model for each task based on performance, cost, and privacy requirements — all through a single API.
Stripe plans to integrate OpenRouter's model routing and billing capabilities into its own infrastructure, effectively creating what Forbes described as "the ledger of AI" — a unified layer for metering, billing, and paying for AI model usage. The deal reflects a broader trend of traditional tech infrastructure companies racing to own the plumbing of the AI economy. (Bloomberg, TechCrunch, Forbes)
Critical AI Infrastructure Vulnerability Exploited for Cryptomining
CISA added a severe vulnerability in the Ray AI framework (CVE-2025-62593, CVSS 9.4) to its Known Exploited Vulnerabilities catalog on Sunday, giving federal agencies just three days to patch — a deadline that falls on Wednesday, August 20.
The flaw enables remote code execution through a DNS rebinding attack that can be triggered via web browsers. Attackers behind the ShadowRay 2.0 campaign are actively exploiting it to hijack GPU clusters and convert them into self-spreading cryptocurrency mining botnets using XMRig.
Ray is widely used to scale AI and machine learning workloads, making the vulnerability particularly dangerous: compromised clusters don't just lose computing resources to crypto miners — they may also expose sensitive training data and model weights. Security firm Oligo, which discovered the campaign, warned that the botnet is self-replicating and can spread across connected infrastructure. (The Hacker News, Dark Reading)