AI persuasion in elections, digital twins, OpenAI safety, next-gen agentic AI...

MIT Technology Review delivers a comprehensive analysis of how AI-driven persuasion is poised to reshape electoral politics worldwide. Recent peer-reviewed studies show that chatbots powered by large language models can shift voter attitudes far more effectively than traditional political ads, with implications for upcoming US elections and global democracies. The piece highlights that while the EU is proactively regulating AI's persuasive uses in campaigns, the US lags behind, leaving oversight to tech companies and exposing elections to foreign and domestic manipulation. The article calls for coordinated policy, technical standards, and international norms to combat these emerging threats, given that the cost and ease of deploying persuasive AI are dropping precipitously. MIT Technology Review

Computer World explores the double-edged nature of digital twin technology, now supercharged by AI. The article outlines the transformative benefits of personalized medicine, industrial optimization, and research, citing MIT and Oak Ridge's use of digital twins to model the US workforce and assess automation risks. However, it also flags significant concerns: studies show digital twins of individuals are often no more accurate than demographic stereotypes, and public trust erodes when brands use synthetic personas instead of real feedback. Meta’s failed social media experiments and survey data underscore the reputational risks and consumer wariness associated with this technology. Computer World

Computer World details OpenAI’s latest research on training large language models, such as GPT-5, to "confess" when they make errors, hallucinate, or violate instructions. This secondary output—separate from the main answer—could be crucial for high-stakes domains like healthcare, law, and finance, where undetected AI errors carry major risks. The Register adds that the confession mechanism incentivizes honesty by rewarding models for truthful admissions, not just correct answers. Gartner analysts note this could enable organizations to operationalize AI safety, combining confessions with traditional governance frameworks to trigger human review or escalate uncertain outputs. Computer World The Register

The Register reports on comments from Anthropic regarding the increasing sophistication of AI-driven cyberattacks, particularly on blockchain smart contracts. The article warns that as AI agents become more adept at discovering vulnerabilities and automating exploits, organizations must deploy AI-powered defenses to keep pace. Gizmodo supplements this by highlighting the rapid escalation in AI-driven cryptocurrency theft and the contentious debates it has sparked in the security community, underlining the urgency for robust, adaptive countermeasures. The Register Gizmodo

MIT Technology Review analyzes why most enterprises remain stuck in AI experimentation mode, despite record investment. Experts interviewed emphasize the need to reengineer organizations for effective human-AI collaboration, focusing on redesigning workflows, integrating governance from the outset, and empowering business leaders—not just technologists—to identify high-impact AI use cases. Early adopters are demonstrating value by starting with low-risk applications and embedding oversight, pointing to a new blueprint for operational AI maturity. MIT Technology Review

Nature discusses the accelerating adoption of AI tools in scientific research, acknowledging significant time and cost savings. However, concerns are rising about overreliance on AI systems, potential erosion of critical expertise, and the risk that automated tools may shape research agendas or introduce subtle biases into findings. The debate now centers on how to balance AI’s undeniable efficiency with the need for human oversight and methodological rigor in scientific discovery. Nature

IEEE Spectrum features insights from cognitive scientist Melanie Mitchell, who argues that current AI benchmarks often fail to reflect real-world cognitive capabilities. Mitchell draws from developmental and comparative psychology, advocating for more rigorous, experimental protocols that probe not just AI’s successes but its failure modes and generalization limits. She emphasizes the importance of skepticism and replication in AI research, warning against overinterpreting benchmark performance as true intelligence or capability. IEEE Spectrum

Wccftech reports that TSMC is accelerating plans to build advanced packaging facilities in Arizona, responding to unprecedented demand from US-based AI chipmakers like NVIDIA and AMD. This move is seen as critical for maintaining US competitiveness in the global AI hardware race. MacRumors adds that Apple is rumored to be extending its chip manufacturing partnership with Intel to iPhones, not just Macs and iPads, further diversifying its supply chain and reducing reliance on a single foundry. Wccftech MacRumors

Computer World examines Google’s launch of Workspace Studio, a no-code platform enabling employees to build custom AI agents that automate daily tasks across Gmail, Drive, Sheets, and third-party apps. While this democratizes agentic AI, analysts note that only a minority of employees possess the skills to fully leverage these tools, and IT departments will face new management and governance challenges as “agentic sprawl” increases. Computer World

The Hill explores how the narrative around AI and employment is shifting, with workers increasingly accepting AI’s role in transforming—not just eliminating—jobs. The article outlines six career paths considered resilient in the age of automation, emphasizing adaptability, human-centric skills, and the importance of upskilling for the evolving workplace. The Hill

DevOps.com highlights the deployment of agentic AI in observability platforms, where autonomous agents now analyze telemetry data, detect issues, and automatically implement fixes. This shift reduces mean time to resolution (MTTR) and diminishes manual toil for site reliability engineers (SREs), signaling a strategic move toward more self-healing, AI-augmented IT operations. DevOps.com