Google DeepMind has joined with Schmidt Sciences, ARIA, Cooperative AI foundation, and Google.org to announce a $10 million funding initiative for academic research into the safety of multi-agent AI systems. The fund aims to kick-start a new field focused on risks that emerge when millions of AI agents interact online, including scams, prompt injections, and cyberattacks. Google DeepMind’s AGI safety lead Rohin Shah emphasized that industry labs have not yet prioritized this long-term research, and that realistic sandbox simulations are needed to understand emergent behaviors. The partners hope to anticipate dangers before mass deployment of agents begins, which Shah estimates is months away. Shah noted that studying single agents in isolation cannot predict large-scale multi-agent dynamics, while Schmidt Sciences’ James Fox warned that without proactive research, the digital commons could descend into anarchy.
This article from MIT Technology Review Insights discusses the impending surge in AI agent adoption, predicting up to 300% growth in two years. It highlights early applications in customer service, HR, and sales boosting productivity by 30-50%. Leaders face the challenge of redesigning roles, reskilling employees, and maintaining workplace culture. The piece emphasizes the need for governance and a mindset shift to treat agents as collaborators. It also underscores the importance of adapting skills and management strategies for a blended workforce.
At SXSW London 2026, MIT Technology Review’s Will Douglas Heaven presented five major themes shaping AI: generative AI is mundane in office automation but its employment impact is unclear; real-world harms include deepfake-fueled violence and election disruption, dangerous chatbot relationships, and military LLMs advising target selection. Public backlash is mounting through protests, cultural boycott, and data center opposition, with some incidents turning violent. AI for science holds genuine promise, with tools like Google DeepMind’s Co-Scientist and OpenAI’s 2028 automated researcher goal, but risks narrowing research and creating fake results. Finally, AI’s ubiquity creates both exhaustion and hype, with the road ahead looking more like a marathon than a sprint.
Attackers used Meta's AI customer support agent to steal Instagram accounts by simply asking it to change linked email addresses. The exploit was remarkably simple, requiring only a VPN to match the account owner's location. This incident highlights that while much attention is on advanced AI threats like Anthropic's Mythos, more mundane attacks targeting AI agents pose significant risks. Experts emphasize the need for rigorous red-teaming and guardrails before deploying such agents. Companies face a trade-off between agent capability and security, as the pressure to innovate often leads to inadequate precautions.
The global healthcare sector is under strain from underinvestment and staff shortages, with the WHO warning of an 11 million worker shortfall by 2030. Many providers are turning to agentic AI, with 68% already adopting AI agents to automate back-office tasks, collaborate with medical teams, and triage patients. At Hospital for Special Surgery, AI agents handle 1,100 insurance claims per month, reducing appeals time from 45 minutes to 5 minutes and improving success rates. The technology is seen as a general-purpose tool that requires a unified data strategy and human oversight for safety. Proponents believe agentic AI can free clinicians for complex, specialized care, rehumanizing healthcare.