https://x.com/i/grok/share/b57e2fa017df4d68a3e9e296a5256135

"They are running a giant experiment of which we are the guinea pigs" Speaking at the opening of Berlin Science Week, Turing Award laureate and one of the "Godfathers of AI" Yoshua Bengio said that the AI industry is effectively exempt from basic safety rules The scientist pointed out that Anthropic's IPOfilings explicitly warn that advanced AI could pose an existential threat to humanity. "And no one's doing anything! This IPO is going to create hundreds of new billionaires — just what the world needs," Bengio said ironically, as he prepares to open a Berlin office for LawZero, a non-profit lab dedicated to building safe AI. According to Bengio, the development of super-powerful AI cannot be left to competing private companies. Instead, it must be brought under global public governance. @HiAiUAE

AI Agent Performs Brain Surgery on a Mouse Founder of biotech startup Exin Therapeutics, Gabriel Ocana-Santero,shared this development. Scientists gave Sonnet 5 full access to an automated Arduino-based surgical setup, a table of gene therapy vectors, and a single goal: reduce seizures in an epileptic mouse. The agent determined that the most effective approach would be a modification that enhances the release of GABA (the primary inhibitory neurotransmitter), selected the drug and brain region, calculated the necessary coordinates based on three scientific papers, and injected 250 nanoliters of a viral solution carrying the required gene edits into the mouse's brain while monitoring the process via a camera. The AI then stitched the wound and completed the surgery — the mouse is alive and recovering well. The authors have not yet commented on the procedure's efficacy. Throughout the process, a professional human veterinarian monitored the robot and could have stopped the surgery at any moment. Initially, the authors wanted to use the flagship Opus 5.5 model, but it flatly refused to participate in biological work. Opus 5 started drafting a plan but also stopped because it was "worried" that if safety filters triggered mid-surgery and forced an interruption, it would end badly. Sonnet 5, however, completed the surgery from start to finish without any issues. The authors are extremely excited about the results and plan to continue using agents in real-world experiments. @HiAiUAE