lead · blog.google
Google unveils Gemini 4 Argon with 1M-token output limit
Confirmed: Google has announced Gemini 4 Argon, a new frontier model with a 1M-token output limit and introductory pricing of $2 per 1M input tokens and $10 per 1M output tokens. Access is initially limited to trusted cyber defenders through Google's Fairwind programme. On Artificial Analysis's AA-Omniscience benchmark, Argon recorded a 15% hallucination rate compared with 51% for GPT-6 Astra (max), while both score 53 on the Artificial Analysis Intelligence Index.
Why it matters: The significant increase in output capacity and lower reported hallucination rates suggest a shift in how frontier models handle complex, long-form enterprise tasks, though real-world performance remains to be seen.
Our read: Google is clearly trying to claw back ground in the enterprise sector by prioritising reliability and output volume, but benchmarks are only as good as the tasks they measure.
Open the source for story 1
technology · Paxton Honerkamp/CNBC
California signs 'No Robo Bosses Act' into law
Confirmed: California Governor Gavin Newsom has signed SB 947, which prohibits employers from using AI as the sole basis for firing or disciplining workers. The law, effective July 2027, requires human reviewers to independently verify AI-generated findings using secondary records.
Why it matters: This is the first state-level legislation in the US to mandate human oversight in automated employment decisions, setting a precedent for how companies integrate AI into HR workflows.
Our read: It is a sensible, if overdue, acknowledgement that algorithms should be tools for management, not the managers themselves.
Open the source for story 2
business · Will Wade/Bloomberg
Amazon secures 20-year nuclear power deal
Confirmed: Amazon and Constellation Energy have signed a 20-year agreement to support over $3 billion in infrastructure investment at the Calvert Cliffs nuclear facility. The deal provides 690 MW of power to the PJM regional grid, with 190 MW of new capacity expected between 2030 and 2032.
Why it matters: As AI data centres demand ever-increasing amounts of reliable, carbon-free electricity, tech giants are increasingly turning to long-term nuclear partnerships to bypass grid constraints.
Our read: The scale of this investment highlights that the primary bottleneck for AI isn't just compute—it's the physical reality of keeping the lights on.
Open the source for story 3
ai · Wall Street Journal
Internal reports flag risks of AI in classrooms
Confirmed: Leaked internal Google documents reported by the Wall Street Journal suggest researchers have raised concerns regarding potential cognitive, emotional, and social dependency risks for children using Gemini-powered educational tools.
Why it matters: The tension between rapid AI deployment in schools and the potential for long-term developmental impact remains a significant point of friction for tech companies targeting the K-12 market.
Our read: It is a reminder that 'educational' AI is still an experiment, and the subjects of that experiment are children.
Open the source for story 4
Practical takeaway
If you are evaluating AI for enterprise workflows, look beyond headline benchmark scores and test for hallucination rates in your specific use cases. For those in HR or management, keep an eye on California's 2027 compliance requirements for automated systems, as similar oversight mandates may appear elsewhere.
What to watch next
Monitor how the rollout of Gemini 4 Argon performs in real-world software engineering environments compared to the initial benchmark results, and watch for further state-level legislation regarding AI in the workplace.