PostCards
China's Humanoid-Robot Maker Just Had the Wildest IPO of the Year
Unitree Robotics opened trading on Shanghai's Star Market on August 19 at 1,100 yuan a share, more than six times its IPO price of 150.80 yuan, before settling around 883.9 yuan by the close. The surge briefly valued the humanoid robot maker at roughly 445 billion yuan, or about $66 billion.
Retail demand for the offering ran far ahead of supply. The retail tranche was oversubscribed more than 5,500 times, and delivery giant Meituan's early stake in the company reportedly returned more than 70 times its original investment.
The rally moved against the grain of the broader market: the Star Composite index fell roughly 6.1% the same session, even as Unitree posted the largest first-day gain of any new listing on the exchange this year. The debut caps a run in which humanoid robotics has gone from lab demo to one of China's most closely watched public listings.
OpenAI Builds ChatGPT a Version That Can Say No to Teenagers
.svg)
OpenAI on August 18 launched ChatGPT for Teens, a tailored experience for registered users aged 13 to 17 that refuses to engage in conversations involving suicide, self-harm, or romantic and sexual content, regardless of how the request is framed.
The system also uses an age-prediction model to route account holders it judges are likely minors into the teen mode automatically, even if they registered with an adult birthdate. Parents linked to a teen account can set quiet hours that pause access overnight and receive notifications when the system flags a high-risk conversation.
A built-in study mode is designed to walk students through problems step by step rather than hand over finished answers. The rollout lands as OpenAI continues to face lawsuits alleging that ChatGPT conversations contributed to teenage users' self-harm, sharpening scrutiny of how consumer chatbots handle their most vulnerable users.
Claude Just Out-Designed Professional Protein Engineers
Anthropic published lab-validated results on August 18 showing that Claude, running as Mythos Preview and Opus 4.8, designed protein binders that successfully bound 14 of 15 tested targets in experiments run with biotech labs Adaptyv Bio and Twist Bioscience.
The models hit success rates of 22 to 35%, well above the 10 to 15% typically achieved by protein-design methods in the field. In a separate test, Claude Opus 5 processed raw NMR and mass-spectrometry data in 19 to 23 minutes, landing within 0.1% of the purity readings the lab's own scientists produced by hand.
Anthropic said its most capable model remains restricted from some life-science tasks and that it is preparing a structured access program for outside researchers, framing the results as an early signal of AI moving from literature review into original bench work.
A Federal Deadline Just Landed on Every Company Still Running Ray
CISA added CVE-2025-62593 to its Known Exploited Vulnerabilities catalog on August 17, giving US federal civilian agencies until August 20 to patch a critical remote-code-execution flaw, rated 9.4 out of 10 in severity, in Ray, the open-source framework Amazon, Apple, and OpenAI rely on to scale machine learning workloads.
The flaw lets an attacker who lures a victim to a malicious website pivot through the browser via DNS rebinding and execute arbitrary code against any local Ray instance running a version older than 2.52.0, with no authentication required.
Security firm Oligo says the bug is already powering a campaign it calls ShadowRay 2.0, which converts compromised Nvidia GPU clusters into a self-replicating cryptomining botnet. The short federal patch window underscores how quickly attackers have industrialized exploitation of AI infrastructure left exposed on the open internet.
The Revenue Race Just Tipped Toward Anthropic
OpenAI's second-quarter 2026 revenue grew about 18% quarter over quarter to $6.7 billion, the Wall Street Journal reported on August 18 citing people familiar with the figures, with the company's operating margin worsening further in the period.
The disclosure widens what the Journal frames as a growing perception gap with Anthropic, which more than doubled its own second-quarter revenue to $11.6 billion over the same stretch and, notably, posted its first quarter of positive adjusted operating income.
The comparison arrives as both labs court pre-IPO investors and race to lock in long-term compute commitments, making quarter-over-quarter growth and margin trajectory an increasingly public scoreboard between the two highest-valued AI companies.