Google Launches First AI Satellite for Project Suncatcher

Google has launched its first test satellite for Project Suncatcher, a bold effort to bring AI data centers into orbit. The satellite, carried into space on a SpaceX Falcon 9 rocket, is equipped with four Tensor Processing Units designed to run Google's Gemini AI models. Each test session lasts about 15 minutes before the chips need to cool down. Engineers spent months preparing the hardware to withstand radiation and launch vibrations here on Earth.

The project tackles a growing concern: ground-based data centers consume enormous amounts of electricity and water. Satellites in low Earth orbit have a major advantage, potentially accessing up to eight times more solar power than facilities on the ground. Google envisions connecting clusters of satellites using lasers to form a network capable of handling increasingly demanding AI workloads.

Satlyt, a startup co-founded by a former Google and SpaceX product manager, raised $8 million in seed funding to build software that runs AI models directly on satellites. Its approach draws comparisons to VMware and Snowflake, serving as a platform layer for others' satellite hardware. Satlyt's software is scheduled to launch on a SpaceX rocket alongside Google's Project Suncatcher prototype, and the company aims to be live on 20 percent of satellites by the end of the decade.

Other players are also pushing space-based AI forward. Starcloud secured a spot on Firefly's Elytra orbital vehicle to demonstrate its AI data processing payload in lunar orbit, with a NASA-supported mission targeting a 2028 launch. Firefly expects the vehicle to operate around the moon for at least five years, paving the way for future AI infrastructure beyond Earth.

On the security front, Thales expanded its partnership with Google Cloud to protect agentic AI workflows. The integration combines Thales AI Security Fabric with Google Cloud Gemini Enterprise, giving organizations better visibility into how users, AI agents, models, and tools interact. Meanwhile, Elisity joined the Open Secure AI Alliance to help develop open standards for agent identity, least privilege, and containment of AI threats.

In a separate development, the Dutch Institute for Vulnerability Disclosure was hit by an automated AI attack that exploited two zero-day flaws in Zammad. The vulnerabilities, both scoring 9.4 on the CVSS scale, allowed attackers to run code remotely and gain root access within seconds. Zammad has urged users to upgrade to version 7 to address the issue.

On the policy and research front, former FTC chair Lina Khan argued that existing laws are sufficient to regulate major AI companies, emphasizing enforcement over new legislation. A new book by Praveen Asthagiri warns that AI products need more than just a strong model, pointing to data, governance, and testing as equally critical factors. Researchers are also urged to verify all AI-assisted content carefully, as NIH will not consider applications substantially developed by AI.

Key Takeaways

<ul><li>Google launched its first Project Suncatcher test satellite carrying four Tensor Processing Units to run Gemini AI models in orbit</li><li>Satellites in low Earth orbit can access up to eight times more solar power than ground-based data centers</li><li>Satlyt raised $8 million in seed funding to build AI software that runs on satellites without building its own spacecraft</li><li>Starcloud's AI payload will ride to lunar orbit on a Firefly spacecraft no earlier than 2028</li><li>Thales integrated its AI Security Fabric with Google Cloud Gemini Enterprise to protect agentic AI workflows</li><li>Elisity joined the Open Secure AI Alliance to develop standards for agent identity and threat containment</li><li>An AI-powered attack exploited two zero-day vulnerabilities in Zammad, allowing remote code execution and root access</li><li>Former FTC chair Lina Khan says existing laws already cover AI regulation and should be enforced</li><li>Author Praveen Asthagiri found that improving memory architecture reduced context-loss errors by over 95 percent</li><li>Ukraine's AI platform Avengers can identify enemy equipment in video streams in 2.2 seconds with about 70 percent accuracy</li></ul>

Satlyt raises $8M to bring AI software to satellites

Satlyt, co-founded by Rama Afullo, a former Google and SpaceX product manager, raised $8M in seed funding. The company builds software to run AI models on satellites without needing to build its own spacecraft. Its software is set to launch on a SpaceX rocket alongside Google's Project Suncatcher prototype. Satlyt's approach is compared to VMware and Snowflake, acting as a platform for others' satellite hardware. The company aims to be live on 20% of satellites by the end of the decade.

Google tests AI data centers in space with Project Suncatcher

Google launched its first test satellite for Project Suncatcher, an effort to build AI data centers in orbit. The satellite carries four Tensor Processing Units that will run Google's Gemini AI models for 15 minutes at a time before needing to cool down. The test aims to see if data center building blocks can survive the harsh conditions of space. Google engineers spent months testing the hardware against radiation and launch vibrations on Earth. The project addresses growing concerns about data centers using large amounts of electricity on the ground.

Google launches first test satellite for in-space AI data center

Google is set to launch its first test satellite for an in-space AI data center on a SpaceX Falcon 9 rocket from California. The satellite is equipped with advanced AI algorithms and high-performance computing systems powered by solar panels and advanced batteries. The test launch is designed to demonstrate the feasibility of AI data centers in space. The company plans a series of tests to validate the performance and reliability of the technology. If successful, it could lead to a new generation of sustainable space-based data centers.

Google sends AI chips into orbit on SpaceX rocket

Google launched its AI chips into orbit for the first time aboard a SpaceX Falcon 9 rocket as part of the Transporter-18 rideshare mission carrying 130 payloads. The Project Suncatcher pathfinder satellite carries Tensor Processing Units to test how they withstand space radiation and heat. Satellites in low Earth orbit can access up to eight times more solar power than on Earth. Google envisions connecting multiple satellite constellations for increasingly demanding AI tasks. Alphabet, a significant investor in SpaceX, plans to expand the project with two more satellites next year.

Google launches Project Suncatcher toward space AI data centers

Google launched Project Suncatcher, a prototype satellite about the size of a fridge carrying four Tensor Processing Units designed for machine learning. The chips will run Google's Gemma AI model for 15 minutes at a time to answer simple queries while managing heat. The satellite is in a sun-synchronous orbit where its solar panels almost never shade, reducing the need for heavy backup batteries. Google, partnering with Planet, envisions clusters of satellites eventually forming space-based data centers connected by lasers. Next year, the company plans to launch two more satellites to test their connection.

Google sends AI chips into orbit for first space data center test

Google launched a satellite nicknamed MVP carrying four of its TPU AI chips into orbit. The satellite lifted off from California on a SpaceX Falcon 9 rocket. It will test running AI queries using Google's Gemini in space. The project called Project Suncatcher could help solve growing energy and water problems from ground data centers.

Thales and Google Cloud expand agentic AI security partnership

Thales expanded its work with Google Cloud to improve security for agentic AI workflows. The new integration combines Thales AI Security Fabric with Google Cloud Gemini Enterprise. This setup gives companies better visibility and control over how users, AI agents, models, and tools interact.

Starcloud AI payload to ride to lunar orbit on Firefly spacecraft

Starcloud's SC-1L AI data processing payload will fly on Firefly's Elytra orbital vehicle. The NASA-supported mission is set to launch no earlier than 2028. It will demonstrate technologies for future AI data centers in Earth orbit and around the moon. Firefly expects the Elytra vehicle to operate in lunar orbit for at least five years.

Elisity joins Open Secure AI Alliance to boost AI safety standards

Elisity announced it joined the Open Secure AI Alliance to help develop open tools for AI security. The company plans to work on standards for agent identity, least privilege, and containment of AI threats. Elisity's platform uses identity-based microsegmentation to help healthcare and infrastructure organizations stop unauthorized movement.

AI-powered hack exploits Zammad zero-day vulnerabilities

The Dutch Institute for Vulnerability Disclosure was hacked by an automated AI attack on September 21. The attack used two zero-day flaws in Zammad, both scoring 9.4 on the CVSS scale. The vulnerabilities allowed attackers to run code remotely and gain root access in seconds. Zammad versions 6.3.0 to 6.5.4 and 7.0.0 to 7.1.3 are affected, and users are advised to upgrade to version 7.

Generative AI and the Research Record

Generative AI can help researchers review literature, organize ideas, and improve writing. However, AI output sometimes contains inaccurate claims, made-up citations, or data that was changed to look better. Research teams must carefully check all AI-assisted content before submitting grants, dissertations, or manuscripts. NIH has stated it will not consider applications substantially developed by AI. Researchers remain responsible for verifying accuracy and maintaining integrity in the research record.

Eyewitness News Extra: New Quinnipiac Poll on AI

A federal judge blocked a Connecticut law that would have banned Glock-style pistols just before it took effect. A new Connecticut law now prevents ICE agents from detaining people at schools, hospitals, and places of worship. The segment also covers a new Quinnipiac poll on artificial intelligence. WFSB's Eyewitness News Extra features interviews with reporters covering public policy and other current topics.

Former Federal Regulator Says Existing Laws Apply to AI

Lina Khan, who served as chair of the Federal Trade Commission during the Biden administration, argues that current laws are sufficient to regulate major AI companies. She wrote an op-ed titled 'There Are Plenty of Laws on the Books to Check the A.I. Giants. Use Them.' In an NPR interview, Khan discussed how existing legal frameworks can be applied to AI technology. Her position emphasizes enforcement of current rules rather than creating entirely new ones.

Expert Warns AI Products Need More Than a Good Model

A new book by Praveen Asthagiri argues that data, architecture, testing, and governance matter as much as the AI model itself. He says a model that works well in demos can fail with real users due to bad data, slow retrieval, or poor monitoring. His book promotes a system-first approach where the model is just one part of a larger product. He found that improving memory architecture reduced context-loss errors by over 95% in one project. The key lesson is that a product must work reliably for everyone every time.

What Can Ukraine's Battlefield AI Do?

Ukraine's AI-powered computer vision platform called Avengers can detect enemy equipment in video streams and identify individual objects in 2.2 seconds. The system is integrated with the DELTA digital system and can identify about 70% of enemy equipment. Ukraine generates over 50,000 video streams monthly from unmanned systems for training AI models. The government created Brave1, a defense-technology cluster, to connect developers with military needs. AI on the front also supports computer vision, autonomy, edge AI processing, and situational awareness.