AI-generated political advertising has surged across US election coverage ahead of the November midterms. According to the Wesleyan Media Project, at least 164 AI-generated or enhanced political ads have aired, costing just under 80 million dollars. These ads depict politicians in fabricated situations, such as placing Texas Republican Senate nominee Ken Paxton in fictional scenes with his Democratic rival James Talarico. Experts warn that such deepfake content could cause voters to distrust all political information and tune out authentic messaging.
On the security front, a zero-day vulnerability named Plugin4Shell has been discovered affecting four major AI coding agents. The flaw impacts Anthropic's Claude Code, OpenAI's Codex, GitHub Copilot, and Google's Gemini CLI. It compromises Git SHA pinning, which allows attackers to swap plugins during installation without verification. As of September 23, two vendors had released fixes, one deprecated the affected tool, and one disputed the findings.
In education and research, Stanford researchers are examining how AI can extend human capabilities in learning and science. Their work emphasizes that AI literacy requires understanding both what the technology can do and the tradeoffs of depending on it. Meanwhile, Minnesota State University Moorhead is launching a Master of Science in Artificial Intelligence, a 30-credit hybrid program that can be completed in as little as one year, with applications opening for a fall 2027 start.
OpenAI CEO Sam Altman is preparing to pitch global AI standards at a UN Security Council meeting. He describes AI as a double-edged sword and stresses the need for responsible human guidance over its development. Experts have cautioned that unregulated AI could lead to catastrophic outcomes. In related academic developments, the University of Auckland Business School launched Huanui: The AI Initiative, which focuses on placing human judgment at the center of AI development and fostering ethical decision-making across business, government, and community sectors.
Other developments show AI reshaping everyday decisions and industry roles. Visitors to Jackson Hole increasingly rely on AI tools to plan trips, prompting discussion among tourism officials about the benefits and challenges of this trend. Legal AI startup Spellbook hired Meltem Kuran-Berkowitz as chief marketing officer, signaling a push toward growth and public trust. A separate study found that consensus among LLM judges is less reliable than assumed, since the judges often share similar training errors, making them equivalent to only 3.5 independent evaluators in some cases.
Key Takeaways
- AI political ads have cost nearly $80 million across at least 164 spots in US midterm elections, raising concerns about voter trust.
- Plugin4Shell flaw affects four AI coding agents: Anthropic's Claude Code, OpenAI's Codex, GitHub Copilot, and Google's Gemini CLI.
- Stanford researchers emphasize AI literacy means understanding both capabilities and tradeoffs of relying on the technology.
- Minnesota State University Moorhead launches a 30-credit hybrid AI master's program completable in one year, with fall 2027 applications open.
- OpenAI CEO Sam Altman plans to propose global AI standards at a UN Security Council meeting.
- University of Auckland launches Huanui: The AI Initiative, focused on human-centered AI development and ethical choices.
- Jackson Hole tourism officials are weighing the benefits and challenges of visitors using AI for trip planning.
- Spellbook hires Meltem Kuran-Berkowitz as CMO, signaling a focus on growth and public trust in AI tools.
- LLM judge consensus can be misleading because shared training errors reduce effective independence to 3.5 judges out of ten.
- Two of four vendors affected by Plugin4Shell had shipped fixes as of September 23, while one deprecated the tool and one disputed the findings.
AI political ads flood US airwaves ahead of midterm elections
AI-generated political ads have spread across US federal, state and local elections ahead of the November midterms. These ads place politicians in fictitious situations and cost nearly 80 million dollars across at least 164 ads tracked so far. Texas Republican Senate nominee Ken Paxton used an AI ad depicting his Democratic rival James Talarico in a fabricated scene. Experts warn that such deepfake ads could erode public trust in election information.
AI political ads warp reality ahead of US midterms
AI-created political ads are proliferating across US airwaves and social media despite a broader backlash against the technology. One ad showed Democratic candidate James Talarico writing about gender in a Bible with crayons, though he never filmed such a scene. The Wesleyan Media Project found at least 164 AI-generated or enhanced political ads this year, costing just under 80 million dollars. Experts say these ads may cause voters to doubt real content and tune out authentic information.
University of Auckland launches human-centered AI initiative
The University of Auckland Business School launched Huanui: The AI Initiative, led by Professor Alexander Richter. The hub focuses on putting people, purpose and human judgement at the centre of AI development. It aims to support research, teaching, innovation and collaboration while encouraging ethical and critical choices about AI. The initiative also seeks to develop tools for using AI responsibly and to engage with business, government and communities.
Stanford researchers examine AI use in education and science
Stanford researchers are exploring how AI can extend human capabilities in education, science and social sciences. They emphasize that AI literacy means understanding both what the technology can do and the tradeoffs of relying on it. The research comes as AI becomes increasingly capable in many fields. The goal is to weigh the benefits and risks of applying AI in learning and scientific settings.
Jackson Hole visitors turn to AI for trip planning
Visitors to Jackson Hole are increasingly using artificial intelligence to plan their trips, raising concerns among tourism officials and public land managers. Tourism leaders and outdoor brand representatives discussed the trend at Jackson Hole Mountain Resort's Trapper's Restaurant. The shift reflects how AI tools are changing the way people research and organize travel to natural destinations. Officials are weighing the benefits and challenges of this growing reliance on AI.
Spellbook names Meltem Kuran-Berkowitz as new CMO
Legal AI startup Spellbook hired Meltem Kuran-Berkowitz as its chief marketing officer. She brings strong views on how to market AI without being creepy. The move signals Spellbook is focusing on growth and public trust in AI tools.
Minnesota State Moorhead launches AI master's program
Minnesota State University Moorhead is launching a Master of Science in Artificial Intelligence. The 30-credit hybrid program combines online and in-person classes and can be finished in as little as one year. Students can focus on business or healthcare administration. Applications opened for the first cohort starting in fall 2027.
Sam Altman to push global AI standards at UN
OpenAI CEO Sam Altman plans to pitch global AI standards as the safety debate grows stronger. He met with diplomats ahead of a UN Security Council meeting set for Wednesday. Altman calls AI a double-edged sword and says humans must guide its development responsibly. Experts have warned that unregulated AI could lead to catastrophic outcomes.
Plugin4Shell flaw hits four major AI coding agents
Security firm Air Security discovered a zero-click flaw named Plugin4Shell affecting four AI coding agents. The bug touches Anthropic's Claude Code, OpenAI's Codex, GitHub Copilot, and Google's Gemini CLI. It breaks Git SHA pinning, letting attackers swap plugins during install without verification. As of September 23, two vendors had shipped fixes, one deprecated the tool, and one disputed the findings.
LLM judge consensus often hides shared errors
A new study finds that consensus among LLM judges is not as reliable as it seems. The judges often share similar errors because they are trained in similar ways. In a bank of ten judges, shared errors made them as useful as only 3.5 independent judges. In up to 28% of comparisons, ignoring shared errors led to wrong conclusions about which system was better.
Sources
- AI-generated political ads warp reality ahead of U.S. midterms
- AI political ads warp reality ahead of US midterms
- New initiative charts course for human-centred AI
- Stanford researchers weigh AI applications in education, science
- Raising concerns, Jackson Hole's visitors are using artificial intelligence to plan trips
- Legal AI startup Spellbook appoints Meltem Kuran-Berkowitz as CMO
- Minnesota State Moorhead launches master’s program in Artificial Intelligence
- OpenAI CEO Sam Altman to Pitch Global AI Standards as Safety Debate Intensifies
- Plugin4Shell SHA-Pinning Bypass Hits 4 AI Coding Agents
- Agreement Overstates Evidence: Error Dependence in LLM Judge Consensus
Comments
Please log in to post a comment.