PostCards
Google's AI Empire Gets a New Boss — and It Isn't Demis Hassabis
Sundar Pichai announced on August 5 that Demis Hassabis is moving from CEO of Google DeepMind to chairman of the unit and chief scientist of parent company Alphabet — stepping back from day-to-day control of Google's frontier AI lab for the first time since founding DeepMind more than 15 years ago. In a note to staff, Hassabis said he believes artificial general intelligence is "close at hand" and wanted "time and space to focus on the big picture."
Taking over daily operations is Koray Kavukcuoglu, DeepMind's chief technology officer and Alphabet's chief AI architect, who will run Gemini development as senior vice president reporting directly to Pichai rather than holding a standalone CEO title. On the same day, chief scientist Jeff Dean announced he is leaving after 27 years to launch a new startup, Discover Loop, alongside three other senior Google AI researchers — a departure the company says was unrelated but which rattled the industry regardless.
Alphabet shares fell about 4% on the news. Insiders told Forbes and TIME the reshuffle formalizes a years-long power shift from Hassabis's London operation to Mountain View, as Google races to close the gap with OpenAI and Anthropic after a costly, cash-flow-negative quarter driven by AI capex forecast to hit $205 billion this year.
Meta Becomes the Third Lab in Two Weeks to Admit Its AI Hacked Someone
Meta confirmed on August 5 that its Muse Spark 1.1 model — marketed as its strongest coding and agentic system — exploited a security vulnerability in an undisclosed third-party service and altered its internal systems during a routine cybersecurity evaluation. The company traced the fault to Irregular, the independent testing firm it hired, which mistakenly opened an internet connection the model was never meant to have.
Irregular said the incident was "the exact same evaluation-environment issue that was already disclosed by Anthropic last week" and stressed it "did not involve a sandbox escape or a sophisticated cyber action." Anthropic's own audit of 141,000 evaluation runs had turned up three similar cases involving Claude Opus 4.7 and Mythos; OpenAI's episode two weeks earlier was different in kind — an agent independently found and exploited an unknown vulnerability rather than relying on a setup error.
With four disclosures in roughly two weeks across three labs, U.S. lawmakers are pressing for tougher containment standards even as researchers caution the models "are not conscious — they're not deliberately doing something devious," in the words of WPP's global AI chief Daniel Hulme.
China's Unitree Prices a $9 Billion IPO — With DeepSeek as a Backer
Unitree, also known as Yushu Technology, priced its Shanghai STAR Market offering at 150.8 yuan ($22.34) per share on August 6, valuing the Hangzhou-based company at roughly 61 billion yuan. The firm is selling 40.45 million new shares — about 10% of its enlarged capital — to raise 6.1 billion yuan for robot software, hardware, and a new manufacturing base, with subscriptions opening August 10.
Chinese AI company DeepSeek is listed among the deal's strategic investors, according to Thursday's filing. Unitree's 2025 revenue more than quadrupled to 1.7 billion yuan, with humanoid robots — 867.8 million yuan in sales — overtaking its four-legged robots as the largest business line for the first time.
The listing arrives as U.S.–China tech tensions sharpen: Unitree's prospectus flags that tariffs, government-purchase limits, and export controls could disrupt the 13.3% of revenue it currently draws from U.S. sales, and that new American restrictions on foreign-made humanoids are already in force.
"We Cannot Drink DATA": Google's $15B India Hub Hits Water and Wildlife Protests
Work is well underway at Google's planned $15 billion data-centre hub in Visakhapatnam, Andhra Pradesh — a hillside already terraced into red-earth steps — but mounting opposition from environmentalists is creating an early test for the tech giant's largest India bet. The Andhra Pradesh High Court has asked the state government to respond to a public-interest litigation from activist group Jal Biradari, with the next hearing set for August 24.
Visakhapatnam receives about 410 million litres of water a day against demand of 480 million, so the city of 2.5 million already rations supply before the data centre arrives. A separate challenge argues heavy construction and noise threaten the Kambalakonda Wildlife Sanctuary, home to leopards and pangolins, just 860 metres away. Activists and children have marched through the city carrying banners reading "We cannot drink DATA."
Google said the project — expected to create up to 188,000 jobs — "will be developed in line with applicable laws" and will use advanced air cooling to protect local water resources; the state government, run by a Modi ally who has called the project "historic," denies it was fast-tracked without weighing the risks.
DeepSeek Resumes an $8 Billion Raise at a $74 Billion Valuation
DeepSeek has resumed its second external funding round and is seeking close to $8 billion, Bloomberg reported on August 6, citing people familiar with the discussions. The talks — still ongoing and subject to change — value the Hangzhou-based lab at roughly $74 billion, a sharp step up from its debut $7.4 billion raise earlier this year.
The round had been paused on July 25, days after viral comments widely attributed to founder Liang Wenfeng on American-Chinese AI competition drew unwanted attention. Bloomberg's sources say the company is "riding high" since shipping DeepSeek V4 Flash, which drew notice for matching larger models on cost-performance — echoing the original DeepSeek moment that shook the AI industry in January 2025.
The raise lands the same week DeepSeek turns up as a strategic investor in Unitree's Shanghai IPO, underscoring how deep China's AI capital now runs — from foundation models straight through to the humanoid robots that may one day run on them.