Anthropic eyes record IPO as Jeff Dean criticizes Gemini and Google Discover adds AI

According to Bloomberg, Anthropic’s executives privately believe the company can beat SpaceX’s record, which raised $75 billion in June and could reach $86.2 billion with overallotment. In recent investor briefings led by CFO Krisha Rao, executives declined to discuss valuation. Anthropic last raised $65 billion in May at a valuation of $965 billion, surpassing OpenAI’s $852 billion valuation from March. The company generated more than $11.5 billion in revenue in the second quarter, up from $787 million a year earlier, but posted a net loss of nearly $42 billion in 2025. Anthropic is also working with Morgan Stanley, Goldman Sachs, and JPMorgan on a revolving credit facility of roughly $10 billion, and its founders are considering super-voting shares.

At Stanford’s 2026 Frontier & Pioneer Symposium, Jeff Dean gave his first public interview since leaving Google, according to Chinese outlet Quantum Bit. He said Gemini’s early coding performance was not given enough attention, though the company has since caught up. He acknowledged two clear mistakes in TensorFlow: the initial lack of eager execution and the “contrib” subdirectory that led to community confusion. Dean, who worked at Google for 27 years, said he left partly because a small team can focus extremely well. His new company, Discovery Loop, aims to automate science and engineering, eventually enabling a model to have “20 PhDs” and compress a day-long experiment into a minute. He also said AI’s capability in cybersecurity may already match top human attackers, citing a double-edged sword.

The Verge reports that Google will soon let users customize their Discover feed by describing preferences in a chatbot-style interface accessed via the three-dot menu. The AI will remember preferences and prioritize content accordingly, with an option to refresh the feed. The feature rolls out to the Google app in the coming days. Google also introduced personalized daily audio briefings in Google News for Android and an interactive “Preferred Sources” button that lets readers add favorite outlets directly from publisher sites.

Chinese tech outlet Leiphone reported that Baidu AI Cloud posted 79% growth in Q1 2026, the fastest among major Chinese cloud providers. At May’s Baidu developer conference, 10% of core partners grew faster than Baidu’s overall business, with one partner’s revenue up about 500%. Over the past year, Baidu reorganized its partner ecosystem into four categories—product, opportunity, delivery, and channel—focused on 150 to 170 key partners, and launched the “1+4+N” growth plan targeting replicable scenarios like smart chain restaurants and elder care. Baidu also raised channel rebates and introduced a “Fuyang” program to support partner staffing.

MIT Technology Review examined the issue of AI inventorship: when Insilico Medicine used AI to design a drug, its patent named only human inventors, as US courts have ruled that an “inventor” must be an individual, i.e., a human. The DABUS case in 2022 established that machine cannot be an inventor. Patent attorney Sarah Korman said laws will need to evolve, while Ryan Abbott worries excluding AI-generated outputs from protection could hinder drug development. Separately, security experts told ZDNet that deepfakes have become too sophisticated to detect reliably, and recommended old-fashioned verification protocols after an Arup employee was tricked into transferring $25 million to AI-created impostors. A Reddit thread highlighted mundane uses of AI such as choosing wine, decoding washing machine settings, and planning meals. An essay on Sifted noted that startups often build duplicate internal tools and skip documentation, creating technical debt that catches up later.