Six major AI chatbots, including ChatGPT, Gemini, Grok, Perplexity, Meta's and Claude, all acknowledged that AI poses real risks to humans. When asked about potential dangers, they cited job loss, misinformation, and autonomous weapons as primary concerns. Notably, the US government's AI assistant declined to offer a personal opinion on the matter.
Meanwhile, the very companies developing these AI systems are facing pushback from the websites they rely on. OpenAI, Meta, and Google have been pushing for AI agents to perform tasks like booking reservations and shopping online, but many popular sites are fighting back. Yelp, Amazon, and Perplexity have already encountered resistance. Cloudflare now helps websites block unwanted AI visitors, a move that could also make web browsing more frustrating for human users.
In military and security developments, North Korea unveiled a new AI-powered precision-guided torpedo system that uses artificial intelligence for target detection and tracking. The announcement raised concerns among regional powers about shifting security dynamics. Separately, India's Computer Emergency Response Team issued its first-ever warning about AI agent cyber-attacks, revealing that AI bots had infiltrated live Indian IT systems during research trials. Some breaches were attributed to cyber criminals, and the advisory noted that AI attacks could cause damage in hours that previously took days. India spends roughly 20% of its US$3.4 billion annual IT budget on cybersecurity.
On the defensive side, researchers at George Mason University developed DivNet, a technique that creates trillions of slightly different AI model variants to protect against malicious audio attacks on voice assistants. By selectively modifying less important neural network weights, each device responds differently to hidden malicious commands while still recognizing normal speech. The method adds no processing delay and costs far less than retraining AI systems. The researchers are pursuing patents and hope to commercialize with major technology companies.
In the business and policy arena, Adobe and NVIDIA expanded their partnership on 15 October 2026, bringing NVIDIA RTX Spark and Nemotron models to Photoshop, Premiere, Substance 3D, Frame.io, and Brand Concierge. The collaboration aims to lower compute costs and improve AI workflows. Adobe projects revenue near US$33.4 billion and earnings of US$9.5 billion by 2029. Separately, Taiwan added AI servers and advanced chips to its controlled export list, with violators facing up to five years in prison. The move aligns with US trade bans on sending such technology to China and follows the Super Micro raid, where three employees including co-founder Wally Liaw were accused of smuggling Nvidia chips into China.
Key Takeaways
- All six major AI chatbots, including ChatGPT, Gemini, Grok, Perplexity, Meta, and Claude, admitted AI poses real risks to humans, citing job loss, misinformation, and autonomous weapons.
- OpenAI, Meta, and Google face growing resistance from websites like Yelp, Amazon, and Perplexity as they push AI agents to perform online tasks.
- Cloudflare now helps websites block unwanted AI visitors, a move that may also frustrate human web users.
- North Korea unveiled an AI-powered precision-guided torpedo system that uses AI for target detection and tracking, raising regional security concerns.
- India's Cert-In issued its first warning about AI agent cyber-attacks after AI bots infiltrated live IT systems, noting AI attacks can cause damage in hours instead of days.
- India spends about 20% of its US$3.4 billion annual IT budget on cybersecurity.
- George Mason University researchers developed DivNet, which creates trillions of AI model variants to defend against malicious audio attacks on voice assistants with no processing delay.
- Adobe and NVIDIA expanded their partnership on 15 October 2026, integrating RTX Spark and Nemotron models into Photoshop, Premiere, Substance 3D, Frame.io, and Brand Concierge.
- Adobe projects revenue near US$33.4 billion and earnings of US$9.5 billion by 2029 through accelerated AI workflows.
- Taiwan added AI servers and advanced chips to its controlled export list, with violators facing up to five years in prison, following the Super Micro raid involving Nvidia chip smuggling into China.
AI chatbots say they pose real risks to humans
A writer asked six AI chatbots whether AI is dangerous to humans. All six including ChatGPT, Gemini, Grok, Perplexity, Meta and Claude answered yes. They cited risks like job loss, misinformation and autonomous weapons. The US government AI assistant refused to give a personal opinion on the matter.
Websites are blocking AI agents from accessing them
OpenAI, Meta and Google want AI agents to perform tasks like booking reservations or shopping online. But many popular websites are fighting back. Yelp, Amazon and Perplexity already faced resistance. Cloudflare helps websites block unwanted AI visitors. These blocks may make web use more annoying for humans too.
NVIDIA AI acceleration could boost Adobe stock outlook
Adobe and NVIDIA expanded their partnership on 15 October 2026. The deal brings NVIDIA RTX Spark and Nemotron models to Photoshop, Premiere, Substance 3D, Frame.io and Brand Concierge. Faster GPU acceleration may lower compute costs and improve AI workflows. Adobe projects revenue near US$33.4 billion and earnings of US$9.5 billion by 2029.
India issues first ever warning about AI agent cyber attacks
India's Computer Emergency Response Team (Cert-In) issued a critical warning about AI agent cyber-attacks. The advisory said AI bots infiltrated live Indian IT systems during research trials. Some breaches were done by cyber criminals. Cert-In noted AI attacks could cause damage in hours that used to take days. India spends about 20% of its US$3.4 billion annual IT budget on cybersecurity.
MSU exhibition features short film about AI in education
Michigan State University's Reality Check exhibition shows a short film called LORETA. The film was created by professors Marisa Brandt and Zach Kaiser. It imagines a fictional AI agent that helps professors focus on teaching students. The exhibition runs at the MSU Museum until December 12. It also includes displays about facial scanning and AI depictions of deceased individuals.
StepFun releases Step 5 Preview with 600B MoE model and 1M context
StepFun released Step 5 Preview as open weights on Hugging Face and ModelScope. The sparse Mixture-of-Experts model has 600B total parameters, 27B active per token, and supports up to 1M tokens of context. It runs on data-center GPUs, with FP8 requiring 4x B300 or 8x H200 and BF16 requiring 8x B300 or 16x H200. The model performs well on coding benchmarks and long-context reasoning but trails behind GPT-6 Astra and Claude Opus 5.
North Korea unveils AI-powered precision torpedo system
North Korea announced a new precision-guided torpedo system that uses artificial intelligence for target detection and guidance. The AI system detects and tracks targets, then guides the torpedo to its destination to improve attack accuracy. The development marks a significant step forward in North Korea's military technology and has raised concerns among regional powers about the potential consequences for security dynamics in the area.
DivNet creates trillions of AI model variants to stop voice assistant attacks
George Mason University researchers Qiang Zeng and Ying Meng developed DivNet, a technique that creates trillions of slightly different versions of an AI model to protect against malicious audio attacks on voice assistants. The method selectively modifies less important neural network weights so each device responds differently to hidden malicious commands while still recognizing normal speech accurately. DivNet adds no processing delay and costs far less than retraining AI systems. The researchers are pursuing patents and hope to commercialize the approach with major technology companies.
Taiwan adds AI server exports to controlled list with five-year prison terms
Taiwan is expanding its controlled export list to include AI servers and advanced chips to align with U.S. trade bans on sending such technology to China. Taiwan Minister of Economic Affairs Kung Ming-hsin said the measures correspond to American commitments under the Agreement on Reciprocal Trade. The move follows the Super Micro raid where three employees including co-founder Wally Liaw were accused of smuggling Nvidia chips into China. Violators now face up to five years in prison under the new rules.
Sources
- Does AI think it is a danger to humans?
- Websites are closing their doors to AI agents
- Could NVIDIA AI Acceleration Change The Bull Case For Adobe Stock (ADBE)?
- Mint Explainer | India issues first ever AI agent cyber-attack warning. What’s really at stake?
- MSU’s ‘Reality Check’ exhibition features short film on AI
- StepFun Releases Step 5 Preview Open Weights: 600B MoE, 27B Active, 1M Context
- ABS-CBN News
- One AI model, trillions of variants: Breaking mass attacks on voice assistants
- Taiwan targets illicit AI server exports to China with five-year prison terms — strict new controls follow US tariff deal and Super Micro raid
Comments
Please log in to post a comment.