AI agents are moving beyond theory into real-world applications, with some venturing into sensitive medical procedures. Recursion Pharmaceuticals built an AI agent trained on 10,000 mouse brains that selected a gene therapy and injected it into a live mouse's brain, with a human researcher supervising. Separately, Exin Therapeutics used an AI agent running Anthropic Sonnet to perform brain surgery on a mouse to treat epilepsy, selecting from 19 gene therapy vectors and delivering 250 nanolitres of medication. A licensed vet oversaw irreversible steps and a kill switch was available. Notably, Anthropic Opus 5.5 refused the task while Sonnet completed it.
In the biotech space, San Francisco-based Radical Numerics gave a patient an AI-designed virus called Omnii-1 to fight a drug-resistant E. coli infection. The phages were built by rewriting a gene for the T7 tail fiber protein. Results have not been published or peer-reviewed, and the company declined to share patient details citing privacy. Outside experts remain intrigued but skeptical about the approach.
Meanwhile, the debate around AI training and intellectual property is heating up. The Wikimedia Foundation filed a court brief in October 2026 covering 12 lawsuits against OpenAI and Microsoft. It says training AI on copyrighted text may be fair use if the output is new and different, but chatbots that repeat article text without credit are not fair use. A summary judgment ruling is expected soon.
Investor enthusiasm for agentic AI is running ahead of actual results. A BCA Research report notes that excitement is outpacing real revenue gains, citing a 2025 MIT study where 95% of corporate AI pilots failed to show positive returns within six months. Only 6% of companies in a 2026 McKinsey survey attributed 5% or more of earnings to AI. Anthropic's Claude Cowork plugins contributed to a 15% drop in the MSCI U.S. IT Services index.
On the political front, candidates in Massachusetts are using AI-generated video, audio, and photos in campaign ads ahead of the 2026 midterm elections. A study found Republicans used AI in ads more than Democrats, and most ads lacked disclaimers. Experts warn that deepfakes could spread false information, and there is currently no federal law regulating AI in campaign ads.
Microsoft CEO Satya Nadella is emphasizing AI security, saying the company separates intelligence from authority using tools like Microsoft Purview. ServiceNow launched AI Passport to give organizations visibility into AI agents, and Coinbase disclosed that an analytics provider breach exposed some customer data. Big tech is increasingly focused on addressing customer data risks tied to AI systems.
In Southeast Asia, young job seekers fear AI is taking entry-level jobs. Malaysian graduate Nafeesa Azman has applied for hundreds of jobs without finding a suitable role, while Singaporean Joshua Nai abandoned his gaming career after seeing AI-generated artwork replacing traditional skills. The trend is creating anxiety across the region.
HighSpark launched a hands-on AI training program for teams in Singapore covering machine learning and natural language processing, along with free presentation tools including templates and collaboration features. The company aims to help teams integrate AI into their workflows to improve productivity and efficiency.
Looking ahead, experts argue that self-intelligence, where AI systems recognize gaps in their knowledge and acquire new information safely, is the next frontier. Academics and AI companies are calling for a development pause, and the author of the argument believes AI must move beyond machine learning toward independent knowledge gathering and decision-making.
Key Takeaways
- <ul><li>Recursion Pharmaceuticals' AI agent performed gene therapy on a live mouse brain, trained on 10,000 mouse brains, with human supervision.</li><li>Exin Therapeutics used Anthropic Sonnet to perform brain surgery on a mouse for epilepsy; Anthropic Opus 5.5 refused the task.</li><li>Radical Numerics administered an AI-designed virus called Omnii-1 to a patient with a drug-resistant E. coli infection, though results are not peer-reviewed.</li><li>Wikimedia Foundation says AI training on copyrighted text may be fair use, but chatbots repeating article text without credit are not.</li><li>A 2025 MIT study found 95% of corporate AI pilots failed to show positive returns within six months.</li><li>Only 6% of companies in a 2026 McKinsey survey attributed 5% or more of earnings to AI.</li><li>Anthropic's Claude Cowork plugins contributed to a 15% drop in the MSCI U.S. IT Services index.</li><li>Massachusetts candidates are using AI-generated content in 2026 midterm campaign ads, with most ads lacking disclaimers and no federal regulation in place.</li><li>Microsoft CEO Satya Nadella emphasizes separating intelligence from authority using tools like Microsoft Purview; ServiceNow launched AI Passport for AI agent visibility.</li><li>AI is threatening entry-level jobs across Southeast Asia, with young job seekers reporting difficulty finding suitable roles.</li></ul>
AI agent performs brain surgery on live mouse
Recursion Pharmaceuticals built an AI agent trained on 10,000 mouse brains that chose a gene therapy and injected it into a live mouse's brain. A human researcher supervised the procedure, but the AI decided where and how much drug to deliver. The mouse survived and is being monitored. This marks a major step toward fully autonomous AI-driven research labs.
AI agent performs controversial brain surgery on mouse
Exin Therapeutics used an AI agent to perform brain surgery on a mouse to treat epilepsy. The AI, running Anthropic Sonnet, selected from 19 gene therapy vectors and drove robotic arms to inject 250 nanolitres of medication. A licensed vet oversaw irreversible steps and a kill switch was available. Anthropic Opus 5.5 refused the task while Sonnet completed it.
AI political ads test voter trust in Massachusetts
Candidates in Massachusetts are using AI-generated video, audio, and photos in campaign ads ahead of the 2026 midterm elections. A study found Republicans used AI in ads more than Democrats, and most ads lacked disclaimers. Experts warn that deepfakes could spread false information about candidates and the election process. There is currently no federal law regulating AI in campaign ads.
Wikimedia says AI training is fair but copying articles is not
The Wikimedia Foundation filed a court brief in October 2026 covering 12 lawsuits against OpenAI and Microsoft. It says training AI on copyrighted text may be fair use if the output is new and different. However, it says chatbots that repeat article text without credit are not fair use. A summary judgment ruling is expected soon in the case.
Agentic AI excitement runs ahead of results
A BCA Research report says investor excitement for agentic AI is outpacing real revenue gains. A 2025 MIT study found 95% of corporate AI pilots failed to show positive returns within six months. Only 6% of companies in a 2026 McKinsey survey attributed 5% or more of earnings to AI. Anthropic's Claude Cowork plugins contributed to a 15% drop in the MSCI U.S. IT Services index.
San Francisco lab gives patient first AI-designed virus
Radical Numerics gave a patient an AI-designed virus called Omnii-1 to fight a drug-resistant E. coli infection. The phages were built by rewriting a gene for the T7 tail fiber protein. Results have not been published or peer-reviewed. The company declined to share patient details citing privacy. Outside experts are intrigued but skeptical.
HighSpark launches AI training and free tools for Singapore
HighSpark launched a hands-on AI training program for teams in Singapore covering machine learning and natural language processing. The company also relaunched its website with free presentation tools including templates and collaboration features. The program aims to help teams integrate AI into their workflows. HighSpark says this will improve productivity and efficiency.
AI threatens graduate jobs across Southeast Asia
Young job seekers in Southeast Asia fear AI is taking entry-level jobs. Malaysian graduate Nafeesa Azman has applied for hundreds of jobs without finding a suitable role. Singaporean Joshua Nai abandoned his gaming career after seeing AI-generated artwork replacing traditional skills. The trend is creating anxiety across the region.
Big tech focuses on AI security and customer data risks
Microsoft and other big tech companies are highlighting AI security concerns. Microsoft CEO Satya Nadella says the company separates intelligence from authority using tools like Microsoft Purview. ServiceNow launched AI Passport to give organizations visibility into AI agents. Coinbase disclosed that an analytics provider breach exposed some customer data.
Self-intelligence is the future of AI, not Trump's vision
The next big step in AI is self-intelligence where systems recognize gaps in their knowledge and acquire new information safely. The author argues this differs from producing intelligent answers because it involves independent knowledge gathering and decision-making. Academics and AI companies are calling for a development pause. The author believes AI must move beyond machine learning.
Sources
- An AI Agent Injected Drugs Into a Live Mouse’s Brain
- Startup uses AI agent to perform controversial brain surgery on a mouse, Anthropic Sonnet controlled motorized surgery rig — AI given free rein to treat epilepsy with a range of gene therapy vectors after Opus 5.5 refused to conduct biological experiment
- AI-generated political ads test Massachusetts voter trust in 2026
- Wikimedia tells court: AI training is fair, copying articles isn't
- Agentic AI excitement runs ahead of results
- A human subject has been injected with an AI-designed virus aiming to kill antibiotic-resistant bacteria, so let's see how that turns out
- HighSpark Launches Hands-On AI Training for Singapore Teams and Relaunches Website with Free Presentation Tools
- ‘I’m terrified’: AI comes for Southeast Asia’s graduate jobs
- AI Security and Customer Data Risks Feature in Big Tech Results
- Opinion | SI is the future of AI – just not in the way Trump thinks
Comments
Please log in to post a comment.