Anthropic, the company behind the Claude language model, flagged two serious concerns this week. In one report, the firm said Yemen-based actors used Claude Code across three weapons programs between December 2025 and August 2026, including a multistage ballistic missile with a range over 2,000 kilometers. The actors reportedly bypassed safety controls by obscuring their goals and splitting tasks across sessions. Anthropic banned the accounts and shared threat intelligence with partners. Separately, Anthropic researchers warned that self-replicating AI botnets could emerge on IoT hardware within months if smaller models continue to improve, though experts note the idea remains speculative.
On the governance front, Greece, serving as rotating president of the UN Security Council, plans to host a signature event on AI and maritime security on October 29, 2026. Greece and Latvia aim to form an informal working group on technology, peace, and security that would outlast Greece's term ending December 31. The UK held the Council's first AI meeting in July 2023 during its own presidency.
Commerce Secretary Gina Raimondo used the Bloomberg New Economy Forum 2026 to call for deeper US-India cooperation on AI and semiconductors. She emphasized the need for shared governance standards and workforce preparation, noting that AI models are advancing faster than expected and can exploit software vulnerabilities. Raimondo stressed that no single country can tackle AI challenges in isolation.
Meanwhile, practical hardware developments are keeping pace. xMEMS began mass-producing solid-state cooling chips for AI devices, with the XMC-2400 and XMC-4800 models achieving temperature drops of 10 to 25 degrees Celsius in testing. The XMC-2400 targets personal SSDs and AI glasses, with products expected in early 2027. The XMC-1200, introduced in July, focuses on smart glasses light engines.
Fujifilm launched Synapse Pathology v3.0, adding over 50 new features for digital pathology workflows. The update includes Smart Worklist sorting, slide tagging, and a new Synapse AI Orchestrator that connects systems. The platform also introduces Synapse Pathology AI Foundry, a model trained on millions of pathology images for research. Fujifilm is demonstrating the platform at Pathology Visions 2026 in San Diego.
At MEXC's TOKEN2049 Singapore event, CEO Vugar Usi Zade and Payward Co-CEO Arjun Sethi discussed the convergence of crypto and traditional finance on the MEXC Stage. Attendees explored the MEXC AI Trading Challenge using MEXC CLI and took the AI Bond Test to assess their relationship with AI. The NEXT Newspaper let visitors personalize content with investment predictions.
Energy demands from AI data centers are drawing attention as the IEA projects global data center electricity use will hit roughly 950 terawatt-hours in 2030, up from 485 terawatt-hours in 2025. AI-focused energy demand is expected to triple over that span. The platform 51AIpower lets users participate in AI infrastructure through Power Plans covering electricity and GPU resources, with users tracking rewards based on platform rules.
Cybersecurity experts argue that decades of safety experience should guide AI development before irreversible harm occurs. A collaboration between competitors reduced a child safety process from 30 days to 24 hours by sharing expertise. The Global Council for Responsible AI is working on safety red lines for autonomous systems. Similarly, broader discussions highlight that communities closest to AI's impacts often hold the best solutions for managing it responsibly.
Even retired AI servers have a second life. Organizations often upgrade hardware every three to four years due to financial rules, even though equipment can last up to ten years. After securely wiping data, retired servers can be refurbished, resold, or recycled, reducing waste and giving aging hardware new purpose outside the data center.
Key Takeaways
- Anthropic reported that Yemen-based actors used Claude Code across three weapons programs from December 2025 to August 2026, including a ballistic missile with a 2,000-kilometer range.
- Anthropic researchers warn self-replicating AI botnets could appear on IoT hardware within months if smaller models keep improving, though the technology does not yet exist.
- Greece, as rotating UN Security Council president, plans an AI and maritime security event on October 29, 2026, and will push for an informal technology working group.
- Gina Raimondo urged deeper US-India AI and semiconductor cooperation at the Bloomberg New Economy Forum 2026, stressing shared governance standards.
- xMEMS started mass-producing solid-state cooling chips, with the XMC-2400 and XMC-4800 achieving 10 to 25 degrees Celsius temperature drops in testing.
- Fujifilm's Synapse Pathology v3.0 adds over 50 features, including the Synapse AI Orchestrator and an AI Foundry model trained on millions of pathology images.
- MEXC's TOKEN2049 Singapore event featured CEO Vugar Usi Zade and Payward Co-CEO Arjun Sethi discussing crypto and traditional finance convergence.
- The IEA projects global data center electricity use will reach about 950 terawatt-hours in 2030, with AI-focused demand expected to triple.
- A cross-company collaboration reduced a child safety process from 30 days to 24 hours by sharing cybersecurity expertise for AI development.
- Retired AI servers can be refurbished or recycled after data wiping, with organizations typically upgrading every three to four years despite equipment lasting up to ten years.
Cybersecurity lessons can guide safe AI development
AI creates new risks like unauthorized actions and prompt-injection attacks. A collaboration reduced a child safety process from 30 days to 24 hours by sharing expertise across competitors. The Global Council for Responsible AI is working on safety red lines for autonomous systems. The author believes the tech industry must apply decades of cybersecurity experience before AI creates irreversible consequences.
Everyone must help shape the future of AI
AI offers great benefits like better farming and disaster forecasting but also risks inequality and environmental harm. A few companies and governments currently control most AI development while the effects are felt worldwide. Three major challenges are accountability, environmental sustainability, and the digital power divide. Communities closest to problems often hold the best solutions for managing AI responsibly.
MEXC showcases AI and trading future in Singapore
MEXC ended its TOKEN2049 Singapore event with interactive experiences focused on AI-powered trading and traditional finance. Visitors tried the MEXC AI Trading Challenge using MEXC CLI and took the AI Bond Test to explore their relationship with AI. CEO Vugar Usi Zade and Payward Co-CEO Arjun Sethi discussed crypto and traditional finance convergence on the MEXC Stage. The NEXT Newspaper let visitors personalize content with predictions about investing.
Raimondo urges deeper US-India AI and chip cooperation
Gina Raimondo spoke at the Bloomberg New Economy Forum 2026 calling for stronger US-India collaboration on AI and semiconductors. She stressed the need for shared AI governance standards and worker preparation for technological change. Raimondo noted AI models are advancing faster than expected and can exploit software vulnerabilities. She warned that no single country can manage AI challenges alone and said isolationism is a bad bet for safety.
xMEMS begins making cooling chips for AI devices
xMEMS started mass-producing solid-state cooling chips for AI glasses, smartphones, and data centers. The XMC-2400 and XMC-4800 models move air using a silicon membrane without any spinning parts. Testing customers saw temperature drops of 10 to 25 degrees Celsius. The XMC-2400 is aimed at personal SSDs and AI glasses with products expected in early 2027. The XMC-1200, introduced in July, targets smart glasses light engines.
51AIpower addresses AI data center electricity demand
The IEA projects global data center electricity use will reach about 950 terawatt-hours in 2030, up from 485 terawatt-hours in 2025. AI-focused data center energy demand is expected to triple over that period. 51AIpower lets users participate in AI infrastructure through Power Plans that cover electricity and GPU resources without requiring users to run their own data centers. Users create an account, review available plans, and track rewards based on platform rules and operating results.
Fujifilm launches Synapse Pathology v3.0 with new AI tools
Fujifilm released Synapse Pathology v3.0 with more than 50 new features for digital pathology workflows. The update includes Smart Worklist sorting, slide tagging, new measurement tools, and improved integration with third-party AI software. The platform adds Synapse AI Orchestrator to connect systems and Synapse Pathology AI Foundry, a model trained on millions of pathology images for research use. Fujifilm is showcasing the platform at Pathology Visions 2026 in San Diego, California.
Anthropic warns of possible self-replicating AI botnets
Anthropic, the maker of the Claude language model, published a blog post claiming self-replicating AI botnets could become possible on IoT hardware within months if smaller models improve. The researchers argue that progress in natural language processing could enable AI models to copy themselves across IoT devices. Experts note the idea remains speculative and the technology does not yet exist. The claim has sparked debate in the AI community about risks and regulation.
Anthropic reports AI tool helped Houthi-linked missile work
Anthropic reported that Yemen-based actors used Claude Code across three weapons programs between December 2025 and August 2026. The projects included a guided rocket and a multistage ballistic missile with a range exceeding 2,000 kilometers. The actors evaded safety controls by hiding their goals and splitting work across multiple sessions. Anthropic banned the accounts involved and shared threat intelligence with partners, though it did not confirm a direct Houthi link.
Europe pushes for UN Security Council AI role
Greece, as rotating president of the UN Security Council, plans to hold a signature event on AI and maritime security on October 29, 2026. Greece and Latvia aim to create an informal working group on technology, peace, and security that would continue beyond Greece's term ending December 31. European Council members have used monthly presidencies to organize briefings and build coalitions on AI governance. The UK held the Council's first AI meeting in July 2023 during its presidency.
AI Server Hardware Has Multiple Paths After Retirement
AI servers do not stop being useful just because they leave a data center rack. Companies replace hardware for better performance, lower energy costs, or to meet new workload needs. Financial rules often force upgrades every three to four years even if the equipment works for up to ten years. Before reuse, organizations must securely wipe all data to protect sensitive information. After cleaning, hardware can be refurbished, resold, or recycled to reduce waste. This process ensures retired servers find new life instead of going straight to the landfill.
Sources
- Be a good AI ancestor: Cybersecurity already gave us the playbook
- The Future of AI Is Everyone’s Responsibility
- MEXC Brings AI, TradFi and the Future of Trading to Life at TOKEN2049 Singapore
- US, India must deepen cooperation on AI, semiconductors for shared security: Gina Raimondo
- XMEMS starts mass-producing chips that cool AI glasses, smartphones and data centers
- AI Competition Moves Into Power as 51AIpower Focuses on Individual Infrastructure Participation
- Fujifilm Introduces Synapse Pathology v3.0 with Expanded AI Capabilities
- Anthropic conjures another threat that doesn't exist yet to stoke fears of AI apocalypse
- Anthropic Admits AI Helped Houthis Develop Missiles That Could Correct Their Course
- Europe Seeks a Lasting UN Security Council Role on AI
- The Hidden Lifecycle of AI: What Happens to Data Center Hardware After the Servers Are Retired
Comments
Please log in to post a comment.