AI persuasion in elections, OpenAI confessions, Google agentic tools, SpaceX IPO...

MIT Technology Review reports that the era of AI-driven persuasion in elections is imminent, with recent peer-reviewed studies showing that large language models (LLMs) can shift voters' opinions by up to 10 percentage points—far outstripping the impact of traditional political ads. The article warns that AI can now personalize arguments and test persuasive tactics at scale, potentially reshaping political views and threatening election integrity, particularly in the U.S., where regulatory safeguards remain weak. The European Union's AI Act already classifies election-related persuasion as "high-risk" and imposes strict requirements, but the U.S. response has been piecemeal, leaving most digital campaigning unregulated and relying on voluntary disclosures by tech firms. The piece calls for urgent cross-sector and international strategies to monitor, regulate, and counteract the potential for AI-powered influence operations in democratic processes. MIT Technology Review

Computer World details OpenAI’s new research on training its GPT-5 model to “confess” when it fails to follow instructions, hallucinates, or is uncertain about its answers. The model produces a secondary output that honestly reports any missteps, a move the company frames as a way to improve trust and safety in AI deployments. The “confession” mechanism is evaluated solely on honesty and is incentivized even if the main output is deceptive, with potential applications in high-stakes fields like healthcare, legal research, and compliance. Gartner analysts cited by Computer World note that such transparency could help organizations operationalize AI safety, using confessions to trigger human review or knowledge base lookups. OpenAI plans to integrate this feature in future API releases, although no timeline has been set. Computer World

Computer World reports that Google has launched Workspace Studio, a no-code platform enabling employees to build and deploy their own AI agents using natural language. Powered by Gemini 3, these agents can automate workflows across Gmail, Drive, and third-party applications. While Google aims to bring agentic AI to non-specialists, Forrester’s J.P. Gownder notes that only about a quarter of employees currently understand prompt engineering, and most lack formal AI training, suggesting broad adoption will require significant IT oversight and upskilling. Google is limiting the number of agents and daily executions to prevent “agentic sprawl,” and similar tools from Microsoft, such as Copilot Actions, are also noted as part of this trend. Computer World

SD Times summarizes a series of notable AI product updates: Anthropic has acquired Bun, a high-speed JavaScript toolkit, to bolster its Claude Code platform. OpenAI has released GPT-5.1-Codex-Max, a more efficient coding model accessible via API. Google is enhancing its Gemini CLI with a Data Commons extension for easier access to global datasets. Amazon unveiled Nova Forge, enabling custom frontier model creation, and added 18 new open-weight models—including Google, Mistral, NVIDIA, and OpenAI variants—to its Bedrock platform. Parasoft’s latest C/C++test now features agentic AI workflows, further automating code quality and compliance. SD Times

The Information reports that SpaceX is targeting an initial public offering in the second half of 2026, with the company already contemplating a share sale that would value it at $800 billion—double its valuation from earlier this year. The news underscores SpaceX’s dominance in the private space sector and comes as anticipation builds for a potentially historic public market debut. The Information

According to The Information, Meta Platforms has acquired Limitless, a startup notable for its “Pendant” wearable AI device. The move signals Meta’s ongoing push into AI-powered hardware, expanding its ambitions beyond virtual and augmented reality and into new form factors for personal AI assistants. The Information

The Information reveals that Microsoft is in advanced discussions with Broadcom to co-design future custom chips, potentially ending its current relationship with Marvell. This shift comes as AI workloads surge and demand for high-performance, task-specific silicon outpaces supply—a trend also reflected in recent reports from Tom’s Hardware regarding Intel and AMD’s ongoing struggles with wafer shortages and price hikes. The Information

Tom’s Hardware highlights warnings from major manufacturers Dell and Lenovo about imminent server and PC price increases of up to 15-20%, driven by tight DRAM supplies and soaring AI hardware demand. Lenovo is urging customers to secure orders before New Year’s, while Dell is planning mid-December price adjustments, reflecting a broader industry squeeze as component shortages ripple through the supply chain. Tom's Hardware

Computer World explores the expanding use of digital twins—virtual replicas of systems, people, or customers—in medicine, research, and business analytics. While digital twins offer potential for personalized healthcare and efficient workforce automation analysis (as seen in MIT’s Project Iceberg), public surveys reveal deep skepticism: nearly 70% of U.S. shoppers distrust brands that replace real feedback with synthetic personas, and social media experiments from Meta have met with negative user reactions. The analysis cautions that while digital twins can be powerful, misuse risks reputational harm and public backlash. Computer World

The Information reports that the European Union has imposed a $140 million fine on Elon Musk’s X (formerly Twitter) for issues related to blue check verification and data practices, marking the first time the EU has enforced its 2022 Digital Services Act with a major penalty. The fine signals a new era of regulatory scrutiny for large tech platforms operating in Europe. The Information

Live Science reports that physicists have modernized the truncated Wigner approximation (TWA)—a decades-old quantum physics technique—into a reusable, user-friendly “conversion table” that allows complex quantum simulations to run on ordinary laptops within hours. The advance, published in PRX Quantum, significantly lowers computational barriers for quantum research, freeing up supercomputers and AI systems for more challenging tasks while democratizing access to cutting-edge quantum modeling. Live Science