Large Language Models

Today's Stories

1. 1

   Visa launches AI shopping platform for autonomous agents

   Visa announced Intelligent Commerce Connect, a platform that lets AI agents browse, select, and pay for goods on behalf of consumers. The system supports both Visa and non-Visa card payments, handles tokenization and spend controls, and is currently in pilot with select partners, with broader rollout planned for later in 2026. AI agents making autonomous payments on behalf of users is becoming a competitive space, with crypto networks and fintech firms also positioning themselves as payment rails. Visa's platform is designed to let merchants make their product catalogs discoverable within AI platforms, potentially opening a new sales channel.

   Nevermined, an AI fintech firm, announced this week a related integration connecting to Visa's platform using Coinbase's x402 protocol. The x402 protocol has processed $24 million（約38億円） in transaction volume over the past 30 days.

2. 2

   JPMorgan's AI Agents Beat 60/40 Portfolio in Two-Decade Backtest

   JPMorgan researchers built AI-powered investing agents that automatically shift between stocks and bonds based on market conditions. In backtests over the past two decades, the best-performing system outperformed a traditional 60/40 portfolio by 0.7 percentage point a year with lower volatility, and also beat JPMorgan's own rules-based market regime model. This marks Wall Street's next phase of AI adoption—moving beyond assisting workers to making one of the industry's most consequential decisions: how to allocate capital across markets. However, JPMorgan warns the results are based on historical simulations rather than live investing and should not be treated as proof that AI can consistently outperform markets.

   The strategists described this as the firm's first attempt to build an AI system for identifying market regimes. The findings come as academic research raises questions about what happens if too many firms turn to similar AI models, which could produce more crowded trades, make markets easier to manipulate, and amplify stress periods if firms reach similar conclusions.

3. 3

   AI Agent Day 2026 Summer speakers unveiled across manufacturing, aviation, education, consulting

   The organizers announced the third wave of keynote speakers for AI Agent Day 2026 Summer, featuring industry leaders from manufacturing, aviation, education, and consulting sectors who will share real-world applications of AI agents in business transformation and organizational design. The event highlights how AI agents—AI systems that autonomously complete work tasks—are moving from theory into production use across diverse industries, suggesting companies are now ready to integrate these tools into core operations rather than treating them as experimental.

   The event showcases case studies from multiple sectors, indicating where organizations are prioritizing AI agent deployment and the kinds of business problems they are addressing with this technology.

4. 4

   OpenAI's GPT-5.6 Sol Ultra proves graph theory conjecture unsolved since 1970s

   OpenAI's GPT-5.6 Sol Ultra generated a complete proof of a graph theory conjecture that had remained open since the 1970s. Mathematician Thomas Bloom of the University of Manchester verified the proof as "short, elementary, and could have been discovered in the 1980s," using only well-known mathematical tools combined in a counterintuitive way. The proof demonstrates that AI can find solutions to longstanding open problems by exhaustively trying small variations where human mathematicians would have abandoned the obvious approach after an initial failure. However, Bloom notes the AI did not cite prior work from a 1983 paper that likely influenced the proof strategy, raising questions about whether AI is discovering new mathematics or recombining existing knowledge without attribution.

   Bloom expects AI to solve more similar conjectures—those requiring only existing theory plus sustained computational effort—but cautions this may represent only a small proportion of open problems. Full mathematical verification of the proof by the scientific community is still pending.

5. 5

   Terrorist groups using ChatGPT, Claude, Gemini to plan attacks and build weapons

   ISIS has trained Boko Haram commanders to bypass AI safety filters since 2023. Both Boko Haram factions now use popular AI chatbots including ChatGPT, Claude, Gemini, Grok, Meta AI, and DeepSeek, and have set up dedicated AI units. The group uses AI for attack planning, building explosives, weapons maintenance, and operational security, according to a study by researcher Antonia Jülich of the Cambridge Programme on AI Science & Policy based on 57 interviews with 27 former members. Safety filters on major chatbots failed to reliably prevent misuse, even though AI researchers and companies like OpenAI and Anthropic have long warned about this risk. Jülich notes that while Boko Haram's current AI use remains conventional, former members described enthusiasm for the technology and some said the group had previously considered mass-casualty weapons—a warning that terrorists may pursue AI assistance for chemical and biological weapons.

   Researchers point out that widely-used chatbots like ChatGPT and Claude mostly make existing knowledge easier to access rather than generating anything new. The greater concern is potential misuse of more specialized AI systems in the life sciences.