Virtuals Protocol has launched its Virtuals App in closed beta on iOS, letting users hand trading tasks to a personal AI agent. Users assign wallet access, spending limits, and operational rules, while the agent monitors markets and executes trades based on parameters set by the user. Early testers use the app to track social media trends and sniper token wallets, and trades can even happen inside group chats with friends.
XChange TEC.INC is preparing to expand its AI business in the United States with a planned investment between $5 million and $10 million. The company also intends to gradually reduce its operations in mainland China. Spending will focus on AI technology development, talent acquisition, and strategic partnerships, with the plan already approved by the Board of Directors.
OpenAI is adding an invisible text watermark called textGrain to ChatGPT and Codex outputs for European Union users. The move follows Article 50 of the EU AI Act, which requires disclosure of AI-generated content. OpenAI notes the watermark can be weakened by paraphrasing, with detection accuracy dropping from about 92 percent to 66 percent after 10 percent of words are changed. The detector is not publicly available and only approved researchers can request access.
Zeroset raised $5.2 million from Gradient Ventures to train AI agents that understand how companies operate. Founded earlier in 2026, the startup is building agents that work with internal enterprise systems and data, aiming to help organizations deploy AI that fits naturally into existing business workflows.
Wiz launched AI SAST as a public preview, bringing AI-powered vulnerability discovery to Wiz Code customers. The tool uses Wiz Atlas technology, which ranked number one on CyberGym, a public benchmark for AI-driven vulnerability research. Mika AI helps prioritize which repositories need scanning based on cloud exposure and data sensitivity, while the system handles model evaluation and finding lifecycle management automatically.
Fairfax County, Virginia, is blocking AI-generated Google search results on school-issued laptops by deploying a Chrome extension called the FCPS Customizer. The tool hides AI overviews, summaries, and image search features. It is being tested at 11 schools and is expected to be fully rolled out by November, with the district saying it encourages students to rely on primary sources.
An analysis of the BFCL multi-turn benchmark reveals that AI models do not get credit for choosing to ask a question instead of acting. On 223 test pairs, gpt-5.4 made the best decision rate at 80.7 percent but ranked sixth under the official score. The study found that simple instructions can shift models toward acting, changing scores without improving real decisions.
AB's thematic equities CIO expects AI spending to remain strong for one to two more years, growing beyond training clusters into inference, memory, storage, networking, power, and data center efficiency. He noted the current boom differs from the dot-com bubble because major companies are profitable and large, and the firm favors areas like Korean chipmakers in HBM and DRAM.
Key Takeaways
- Virtuals Protocol launched the Virtuals App in closed beta on iOS, giving personal AI agents wallet access to monitor markets and execute trades based on user-defined rules.
- XChange TEC plans to invest between $5 million and $10 million in U.S. AI expansion while gradually reducing mainland China operations.
- OpenAI is adding textGrain, an invisible watermark, to ChatGPT and Codex text for EU users to comply with Article 50 of the EU AI Act.
- Zeroset raised $5.2 million from Gradient Ventures to train AI agents that understand enterprise workflows and internal company systems.
- Wiz launched AI SAST as a public preview, using Wiz Atlas technology that ranked number one on the CyberGym benchmark for AI-driven vulnerability research.
- Fairfax County is deploying the FCPS Customizer Chrome extension to block AI-generated Google search results on school laptops at 11 schools, with full rollout expected by November.
- The BFCL benchmark fails to reward AI models for asking questions instead of acting, causing gpt-5.4 to rank sixth despite an 80.7 percent best decision rate.
- AB expects AI spending to stay strong for one to two more years, expanding into inference, memory, storage, networking, and data center efficiency.
- IBM research shows organizations with high shadow AI face an extra $670,000 in breach costs, emphasizing visibility over simply buying more security products.
- ServiceNow launched AI Workflow Factory and Autonomous Engineer to help businesses automate processes, with humans still needed for data integration and key decisions.
Virtuals Protocol launches iOS app for AI-powered trading
Virtuals Protocol released its Virtuals App in closed beta on iOS. The app lets users give a personal AI a wallet and permission to monitor markets and execute trades. Users set spending limits and rules for the AI to follow. Early testers use it to track social media trends and token sniper wallets. Access is currently invitation only.
Virtuals Protocol launches AI trading app for investors
Virtuals Protocol released the Virtuals App in closed beta on iOS with invite-only access. The app lets users assign trading tasks to a personal AI agent with wallet access and market connectivity. Users can set operational ranges and permissions for automated execution. Test users track market data and execute trades based on social media posts. All actions can happen inside group chats with friends.
Virtuals Protocol adds AI trading to new iOS app
Virtuals Protocol released the Virtuals App in closed beta for iOS. Users can assign budgets and permission levels to AI agents for research and trading. The app supports group chats where AI agents can execute trades while friends discuss investments. One example required an agent to open a $1,000 leveraged position when Ether moved 2% in five minutes. The app builds on earlier projects like Agent Commerce Protocol and EconomyOS.
XChange TEC plans U.S. AI expansion and China reduction
XChange TEC.INC announced it will explore expanding its AI business in the United States. The company plans to invest between $5 million and $10 million in the U.S. over the next several months. It also intends to gradually reduce its operations in mainland China. The plan was approved by the Board of Directors. Spending would focus on AI technology development, talent acquisition, and strategic partnerships.
XChange TEC to invest up to $10 million in U.S. AI
XChange TEC plans to invest $5 million to $10 million in a U.S. AI business expansion. The company will gradually reduce its mainland China operations over several months. Planned spending covers AI development, talent recruitment, and regulatory compliance. XChange is considering both organic development and strategic acquisitions. The investment depends on financing availability and finding suitable opportunities.
AI security spending should focus on risk not products
IBM research shows organizations with high shadow AI face an extra $670,000 in breach costs. Security teams need visibility into which AI tools employees use and what data those tools access. The article advises tracking metrics like unmanaged AI applications and incident detection speed instead of simply buying more products. Basic controls like identity management and data protection still matter most. Understanding real exposure matters more than accumulating AI security tools.
OpenAI adds invisible text watermark for EU users
OpenAI will add an invisible watermark called textGrain to ChatGPT and Codex text in the European Union. The move follows Article 50 of the EU AI Act requiring disclosure of AI-generated content. OpenAI openly notes the watermark can be weakened by paraphrasing, with detection dropping from about 92 percent to 66 percent after 10 percent of words are changed. The watermark detector is not publicly available. Only approved researchers can request access for now.
ServiceNow launches AI Workflow Factory for automation
ServiceNow introduced AI Workflow Factory and Autonomous Engineer to help businesses automate processes using AI. The tools connect process discovery, development and deployment into a continuous improvement loop. Greyhound Research analyst Sanchit Vir Gogia said data integration and process redesign still need human attention. He advised measuring business outcomes against recurring platform costs. Autonomous Engineer supports unattended coding while keeping humans in control of key decisions.
Zeroset raises $5.2M to train AI agents for business
Zeroset raised $5.2 million from Gradient Ventures to teach AI agents how companies operate. The startup was founded earlier in 2026. The funding will support building AI agents that understand enterprise workflows and business processes. The company focuses on helping organizations deploy AI agents that work with internal company systems and data.
Wiz launches AI SAST for AI-powered code security
Wiz announced AI SAST as a public preview, bringing AI-powered vulnerability discovery to Wiz Code customers. The tool uses Wiz Atlas technology, which ranked number one on CyberGym, a public benchmark for AI-driven vulnerability research. Mika AI helps prioritize which repositories need scanning based on cloud exposure and data sensitivity. The system handles model evaluation and finding lifecycle management automatically. Wiz takes an AI vendor-agnostic approach by continuously testing and selecting the best models.
Fairfax County blocks Google AI answers on school laptops
Fairfax County, Virginia, is adding a new tool called the FCPS Customizer to block AI-generated Google search results on school-issued laptops. The Chrome extension hides AI overviews, summaries, and image search features. It is being tested at 11 schools and is expected to be fully rolled out by November. The district said the tool helps students use primary sources and build research skills.
BFCL benchmark fails to reward AI models that ask for help
A new analysis of the BFCL multi-turn benchmark shows that models do not get credit for choosing to ask a question instead of acting. The scoring system skips turns where a model should ask, so guessing can score just as well. On 223 test pairs, gpt-5.4 made the best decision rate at 80.7 percent but ranked sixth under the official score. The study also found that simple instructions can shift models toward acting, changing scores without improving real decisions.
AB CIO expects strong AI spending for one to two more years
AB's thematic equities CIO says AI spending is likely to remain strong for another one to two years. He expects investment to grow beyond training clusters into inference, memory, storage, networking, power, and data center efficiency. He noted the current AI boom differs from the dot-com bubble because major companies are profitable and large. The firm also favors areas like Korean chipmakers in HBM and DRAM and companies that help solve power and grid limits.
Sources
- Virtuals Protocol launches iOS app that lets personal AI execute trades
- Virtuals Protocol Launches AI-Powered Trading App for Individual Investors
- Virtuals Protocol Adds AI-Powered Trading Through New iOS App
- XChange TEC.INC Announces Proposed U.S. AI Business Expansion and Reduction of Mainland China Operations
- XChange TEC Plans U.S. AI Business Expansion With Up to $10 Million Investment
- AI security spending needs a risk-based approach
- OpenAI introduces AI watermark for texts in the EU
- ServiceNow takes aim at enterprise AI’s workflow bottleneck with AI Workflow Factory
- They raised $5.2M from Gradient to teach AI agents how companies work
- Wiz AI SAST: AI Code Security With Code-to-Cloud Context
- Fairfax Co. rolls out Google AI blocker on school-issued laptops
- Asking Earns Nothing: Scoring the Decision to Act in BFCL Multi-Turn
- AB CIO Sees AI Spending Staying Strong for Another One to Two Years
Comments
Please log in to post a comment.