Google's Satellite Runs TPU, Delivers Space Haiku

Key Takeaways

• Google's Project Suncatcher ran a TPU chip on a satellite in orbit using the Gemma model, with plans to launch more satellites in 2027.
• Cognizant formed a 25,000-person OpenAI group to deliver Codex and ChatGPT to clients, with over 21,000 associates already earning OpenAI badges.
• Cybercriminals shared over 2.3 million AI-related posts on illicit forums in one month, including tools for deepfake-based identity fraud.
• Anthropic's Claude Opus 5.5 and OpenAI's GPT-6-Sol achieved 73 percent security scores for AI-generated code, the highest in Veracode's rankings.
• The PC industry is slowing as AI automation reduces demand, with Intel, Nvidia, and AMD pushing new AI-focused PC designs.
• At least 220 AI-generated political ads appeared ahead of the 2026 midterms, costing campaigns over $137 million, with 81 percent sponsored by Republican groups.
• Nate Soares believes there is over a 50 percent chance advanced AI could cause human extinction within ten years and has stopped saving for retirement.
• A DGIST team placed second in the HACK@CHES 2026 AI chip security contest, using AI to detect semiconductor security flaws.
• AI-generated code security improved from 56 percent to 62 percent, but the most secure AI models are also the most expensive.
• Thirty-one states have passed deepfake or AI disclosure laws, but no permanent federal regulation exists for nonintimate AI content.

Google runs AI inference on a TPU chip in space

Google achieved a milestone with Project Suncatcher by sending a command to a prototype satellite in orbit. The satellite booted up a TPU chip, ran a prompt using the Gemma model, and sent back a haiku about being in space. The prototype launched on October 1 aboard a SpaceX mission and carries four Trillium-generation TPU chips. Google plans to test how the chips handle radiation and thermal stress, with a goal of launching more satellites in 2027.

Google's data center satellite goes online in orbit

Google's first data center satellite is now operating in orbit and carries four Google Tensor Processing Units. The satellite can generate up to 1 kW of power but can only run a TPU for about 20 minutes at a time. Engineers tested it by asking the Gemma 4 model to write a haiku about space. The team will now study how radiation and heat affect the chips in real conditions. If successful, two more satellites with advanced chips may launch in 2027.

Cybercriminals trade AI prompts and deepfake tools

A new Flashpoint report finds that cybercriminals are sharing prompts, jailbreak techniques, and access to restricted AI models instead of building their own. Researchers tracked over 2.3 million AI-related posts on illicit forums in one month, with many on Telegram. Nearly 64,000 posts offered tools to bypass identity checks using synthetic video and cloned voices. One case involved a $25.6 million fraud using deepfake video, showing how AI makes existing fraud easier and cheaper.

DGIST team places second in AI chip security contest

A team from DGIST won second place in the HACK@CHES 2026 contest held in the United States. The team, led by Professor Kim Jae-sung, used AI to detect security flaws in semiconductors. Twelve teams from around the world participated in the event organized by IACR and KAIST. The contest focused on developing AI technology to find and prevent security threats in chips.

AI safety expert Nate Soares stops saving for retirement

Nate Soares, president of the Machine Intelligence Research Institute, believes there is over a 50% chance advanced AI could cause human extinction within ten years. He has stopped contributing to his retirement savings and urges governments to halt training of general-purpose AI systems. He supports focusing AI research on narrower applications like cancer treatment and safer self-driving cars. His view is more pessimistic than estimates from other experts like Geoffrey Hinton, who puts the risk at around 10%.

AI political ads surge ahead of 2026 midterms

At least 220 AI-generated or AI-enhanced political ads have appeared this election cycle, costing campaigns over $137 million. The Wesleyan Media Project found that 81 percent of these ads were sponsored by Republican candidates or groups, 12 percent by Democrats, and the rest by independents. Deepfake videos remain a concern, but AI is mostly used to satirize political opponents in exaggerated ways. Thirty-one states have passed laws against deepfakes or requiring AI disclosure, yet no permanent federal regulation exists for nonintimate AI content.

Cognizant creates 25,000-person OpenAI group

Cognizant announced the formation of its OpenAI Frontier Capability Group, bringing together over 25,000 professionals to deliver OpenAI products like Codex and ChatGPT to clients. The company plans to scale OpenAI tools across its own sales and customer success operations first. More than 21,000 associates have already earned over 40,000 OpenAI badges for completing learning tracks on OpenAI technologies. The group will initially focus on AI-native software engineering, application modernisation, and agentic customer operations.

Asian trading mixed amid AI concerns

The article content was unavailable due to a website access error. The original source could not be retrieved, so no details about Asian trading or AI-related concerns could be confirmed. The article link and ID are preserved for reference.

AI-generated code still fails security checks often

A Veracode report found that overall security for AI-generated code improved from 56 percent to 62 percent, while vulnerabilities dropped from 44 percent to 38 percent. Anthropic's Claude Opus 5.5 and OpenAI's GPT-6-Sol lead the rankings at 73 percent security, surpassing the previous high of 68 percent set by GPT-5.5. However, the improvement was mainly driven by cross-site scripting fixes, which are easier for AI to detect. Veracode warns that the most secure models are also the most expensive and recommends combining AI code with independent security verification.

AI is not a revolution yet, says article

The article compares today's AI concerns to the 19th century Luddite movement, when workers feared new machines would take their jobs. The author argues that, like the Luddites, people today can shape how AI technology spreads and is used in society. The piece encourages a national conversation about AI's benefits and risks to ensure it benefits everyone, not just a privileged few. The author believes the future of AI is not yet written and can be decided together.

PC industry struggles as AI changes the market

The PC industry is slowing down because AI-powered computers and software are automating tasks. Fewer people are buying PCs because AI can do many jobs that PCs used to handle. Companies like Intel, Nvidia, and AMD are pushing new AI-focused PC designs. The industry must adapt to stay competitive in this changing market.

Sources

• Google Successfully Runs Inference From A TPU In Space Under Project Suncatcher
• Google's first data center satellite is online and writing poetry
• AI-as-a-service: Cybercriminals are trading prompts and deepfake tools instead of building models
• DGIST Team Wins Second in AI Semiconductor Security Contest
• Why AI safety expert Nate Soares stopped saving for retirement; what does he want governments to do?
• Artificial intelligence usage in political ads
• Cognizant forms 25,000-member OpenAI group to deliver Codex and ChatGPT
• Asian trading is mixed as worries about AI-related issues grow
• AI-generated code is still frequently failing security checks — with Claude and GPT-6 topping the leader board
• AI Isn't a Revolution—Yet
• The PC industry's painful path to an AI future