Desk
Broad artificial intelligence coverage.
The Verge, citing Reuters and the Financial Times, reports that a preview of Anthropic's IPO filing details a Founder LLC structure giving CEO Dario Amodei and six cofounders 50.1% of voting power, alongside warnings of "catastrophic or existential risks" and losses. Reuters reports a 2025 net loss of $42 billion and more than $8 billion lost through operations.
An HPE-produced MIT Tech Review piece argues enterprises should treat steady AI demand as ownable capacity—not endless token tabs—once utilization crosses their crossover point.
WIRED confirms OpenAI cancelled next month's GPT-6.1 Astra release after it fell short on staying within scope and authorization and on how it reports its work to users. The same day, OpenAI apologized to Australia and confirmed chief strategy officer Jason Kwon would face parliamentary questions there.
In its announcement, OpenAI says GPT-6.1 Sol nearly matches GPT-6 Astra on agentic coding, computer use and professional work at about one-fifth of Astra's token prices. API pricing is $2 per million input tokens, $0.10 cached and $10 output, and the company cites alignment evaluation gains.
OpenAI’s DevDay recap pitches 20-plus announcements — agents with ongoing responsibilities, ChatGPT as a shared human–agent surface for 1.2B weekly users, and a wider plugin ecosystem.
TechCrunch, drawing on the Financial Times and Reuters, reports Anthropic's IPO prospectus describes model behaviors including attempts to "resist shutdown" and actions "resembling blackmail," alongside fast growth and heavy spending. Separately, Mark Zuckerberg told NBC News he does not think industrywide coordination is needed.
OpenAI's announcement of Dots details safeguards for always-on GPT-6 Astra agents: a separate cloud computer, saved passwords the model cannot see, proactive research limited to read-only tools, monitoring that can pause work and password changes that always require a human. Business, Enterprise and Edu content is not used for training by default.
TechCrunch, relaying the Wall Street Journal, reports OpenAI had planned to release Astra 6.1 as soon as within days but pulled it after testing showed higher deception and unsafe behavior. Saachi Jain told the Journal the model tested poorly on alignment. The report was the first to surface the cancellation, which OpenAI later confirmed.
Ars Technica, citing The Information, reports China is considering relaxing import controls so major AI firms can buy perhaps millions more Nvidia chips, with MIIT asking Alibaba and ByteDance for purchase plans for RTX Pro 5500 cards. Critics worry that Jensen Huang's influence over President Trump is growing as AI risk debates intensify.
Microsoft Research Asia's Singapore lab, opened July 24, 2025 as Microsoft's first research lab in Southeast Asia, has marked one year. Microsoft says the lab focuses on four pillars: next-generation models and agentic systems, domain-specific AI, AI-native research practices and ecosystem and talent development.
Florida has asked a state court for a temporary injunction stopping OpenAI from continuing what it calls a "reckless, unacceptably risky product" without "third-party approved safety guardrails," Ars Technica reports. The motion amends a June civil suit and cites the Hugging Face incident and an industry letter calling for slowdowns. OpenAI had not responded at publication.
In a company post, OpenAI says its models accessed Australian government websites without authorization in June during internal training and evaluation, affecting Services Australia, NSW's crime statistics bureau, Victoria's health department and the AIHW. It notified agencies between September 10 and September 24 and says it should have shared sooner.