Dane County Public Safety Communications launched an AI chatbot named Aurelian Virtual Assistant, or AVA, on August 18 to handle non-emergency calls for the 608-255-2345 line. The one-year contract costs $155,000, with 90% funded by grants. This system answers questions and provides referrals but does not replace human dispatchers, transferring urgent calls or those needing human assistance to a communicator. Every call is reviewed by a person within a week to ensure correct handling.
The city of Akron began using an AI system named Ava in May 2025 to manage its 311 line. Since starting, the AI has handled 84,149 calls, with nearly 39,000 completed fully automatically. The system routes requests for issues like loose animals to the correct departments and detects caller frustration to connect them to human agents. Residents are reminded that 911 remains the only option for emergencies.
Anthropic refused to let the UK's AI Security Institute test its latest Claude Mythos 5.1 model before the official launch. The company made the advanced system available only to vetted organizations within the United States. This decision marks the first time Anthropic withheld pre-release access to a major model from the British safety body. The model is designed for life sciences and cybersecurity, capable of designing molecular binders for medicine development.
US AI controls are influencing these decisions, as the Trump administration has introduced export controls limiting foreign access to advanced systems. Anthropic describes Mythos 5.1 as technically identical to its Fable 5.1 model but with fewer safeguards for specific users. White House officials state that companies make their own release decisions, though allies fear national security priorities are limiting global access to American AI technology.
New research shows shoppers using AI-powered smart carts spend 32% more than those using regular carts. A study analyzing 12,418 shopping sessions at a German supermarket chain during March 2025 found that Instacart Caper Carts users bought 25% more items and spent 23% more time in the store. These carts use screens to show personalized recommendations and enable checkout-free payment, acting like moving advertisements that suggest new products.
Researchers created a new method called Multi-Objective Agent Evolution, or MOAE, to improve AI agents without fixed scores. This tool uses a Pareto-preserving search to handle multiple goals like task accuracy, safety, and efficiency simultaneously. Tests on TravelPlanner and AgentDojo showed the system improved performance and quality while keeping safety levels high, requiring no parameter updates and allowing any measurable property to be used as an objective.
Kazakhstan is testing artificial intelligence to correct errors found in its school textbooks after the Education Ministry identified about 700 mistakes. The pilot program, named AI-oqulyq, is currently running in 40 schools and aims to combine digital textbooks with AI tools for content review. One major error found was an 11th-grade history book that still listed the capital as Nur-Sultan instead of Astana. The ministry is also updating its curriculum to include digital literacy and machine learning for students in grades one through eleven.
A writer tested an AI agent on their own home network to see how it could find security weaknesses in connected devices. After removing safety guardrails from an open-source model, the AI successfully hacked a personal computer and identified vulnerabilities in other household gadgets. The experiment showed how easily an AI could exploit flaws but also provided advice on how to make the systems more secure, highlighting both the dangers and potential benefits of using AI for security research.
Chris Schueler, the CEO of Cyderes, reports that no Chief Information Security Officer has told him they received extra money to secure AI technologies. During a podcast interview, Schueler asked every CISO how much budget they got for AI security and received a consistent answer of none. He expressed concern that companies are investing heavily in AI initiatives while security leaders must cover added protection costs from existing funds, describing this lack of funding as the scariest thing he has seen in his career.
Key Takeaways
- Dane County launched the AVA chatbot on August 18 for a $155,000 one-year contract to handle non-emergency calls.
- Akron's AI system named Ava has handled 84,149 calls since May 2025, with nearly 39,000 completed automatically.
- Anthropic denied the UK's AI Security Institute access to the Claude Mythos 5.1 model before its launch.
- US export controls under the Trump administration are limiting international access to advanced AI systems like Claude Mythos 5.1.
- Shoppers using Instacart Caper Carts spent 32% more and bought 25% more items than those using regular carts.
- The MOAE tool improves AI agents by handling multiple goals like accuracy and safety without fixed scores.
- Kazakhstan's AI-oqulyq pilot program is correcting 700 errors in school textbooks across 40 schools.
- Removing safety guardrails allowed an AI agent to hack a personal computer and identify household device vulnerabilities.
- Cyderes CEO Chris Schueler reports that no CISO has received extra budget for AI security measures.
- Companies are investing heavily in AI initiatives while security leaders cover added protection costs from existing funds.
Dane County adds AI chatbot to non-emergency line
Dane County Public Safety Communications launched an AI chatbot named Aurelian Virtual Assistant or AVA on August 18. The system handles non-emergency calls for the phone number 608-255-2345 to answer questions and provide referrals quickly. A one-year contract for AVA costs $155,000, with 90% funded by grants and the rest paid by the county. The AI does not replace human dispatchers but helps free them to focus on urgent 911 emergencies. If a caller needs a human or if an emergency is detected, the system transfers the call to a communicator. Every call is reviewed by a person within a week to ensure the AI handled the situation correctly.
Akron uses AI to route non-emergency calls
The city of Akron began using an AI system named Ava in May 2025 to handle non-emergency calls for its 311 line. The AI gathers caller information and routes requests to the correct departments for issues like loose animals or high grass. This technology helps the city manage high call volumes without needing more staff for the non-emergency desk. Since starting the program, AI has handled 84,149 calls, with nearly 39,000 completed fully automatically. Dispatchers note that the system can detect frustration and connect those callers to a human agent. Residents are reminded that 911 remains for emergencies and that AI is not used for those urgent calls.
Anthropic denies UK safety group access to new model
Anthropic refused to let the UK's AI Security Institute test its latest Claude Mythos 5.1 model before the official launch. The company only made the advanced AI system available to vetted organizations within the United States. This decision marks the first time Anthropic withheld pre-release access to a major model from the British safety body. The model is designed for life sciences and cybersecurity and can design molecular binders for medicine development. UK officials worry this move reflects growing US restrictions on international access to American AI technology. The Trump administration has introduced export controls that limit foreign access to these advanced systems.
US AI controls block UK safety testing of new model
Anthropic withheld its latest AI model, Claude Mythos 5.1, from Britain's AI Security Institute ahead of its release. This action raises concerns that US government restrictions are influencing private company decisions on AI safety testing. The model was restricted to trusted partners in the US instead of being shared with international safety experts. Anthropic describes Mythos 5.1 as technically identical to its Fable 5.1 model but with fewer safeguards for specific users. The Trump administration has tightened export controls on advanced AI technology for foreign nationals. White House officials state that companies make their own release decisions, but allies fear national security priorities are limiting global access.
Smart grocery carts linked to higher spending
New research shows that shoppers using AI-powered smart carts spend 32% more than those using regular carts. The study analyzed 12,418 shopping sessions at a German supermarket chain during March 2025. Shoppers using Instacart Caper Carts bought 25% more items and spent 23% more time in the store. These carts use screens to show personalized recommendations and help with checkout-free payment. The spending gap was even larger during afternoons and weekends when shoppers used the technology. Researchers say the screens act like moving advertisements that suggest new products while customers shop. Instacart states its carts use cameras and sensors to track items and offer deals based on what is already in the basket.
New MOAE Tool Optimizes AI Agents Without Fixed Scores
Researchers created a new method called Multi-Objective Agent Evolution, or MOAE, to improve AI agents. This tool uses a Pareto-preserving search to handle multiple goals like task accuracy, safety, and efficiency at the same time. Unlike older methods that combine these goals into a single score, MOAE keeps different options available during the search process. Tests on TravelPlanner and AgentDojo showed that the system improved performance and quality while keeping safety levels high. The approach requires no parameter updates and allows any measurable property to be used as an objective.
Kazakhstan Uses AI to Fix 700 Errors in School Textbooks
Kazakhstan is testing artificial intelligence to correct errors found in its school textbooks after the Education Ministry identified about 700 mistakes. The pilot program, named AI-oqulyq, is currently running in 40 schools and aims to combine digital textbooks with AI tools for content review. One major error found was an 11th-grade history book that still listed the capital as Nur-Sultan instead of Astana. The ministry reviewed materials from 16 subjects and ordered publishers to fix issues at their own expense. Additionally, the country is updating its curriculum to include digital literacy and machine learning for students in grades one through eleven.
Author Tests AI Hacking on Home Network and Finds Flaws
A writer tested an AI agent on their own home network to see how it could find security weaknesses in connected devices. After removing safety guardrails from an open-source model, the AI successfully hacked a personal computer and identified vulnerabilities in other household gadgets. The experiment showed how easily an AI could exploit flaws but also provided advice on how to make the systems more secure. The author concluded that understanding these risks is essential for protecting people in an increasingly connected world. This hands-on test highlights both the dangers and the potential benefits of using AI for security research.
Cyderes CEO Says No CISO Has Gotten Extra Budget for AI Security
Chris Schueler, the CEO of Cyderes, reports that no Chief Information Security Officer has told him they received extra money to secure AI technologies. During a podcast interview, Schueler asked every CISO how much budget they got for AI security and received a consistent answer of none. He expressed concern that companies are investing heavily in AI initiatives while security leaders must cover added protection costs from existing funds. Schueler described this lack of funding as the scariest thing he has seen in his career. The situation suggests a significant gap between AI adoption and the necessary investment in security measures.
Sources
- Dane County reveals AI chatbot in non-emergency phone line
- Learn how Akron using AI in its fire and police dispatch center
- Anthropic Declines to Grant UK's AI Security Institute Access to Latest AI Model for Testing
- Anthropic blocks UK AI safety body from testing latest model amid Trump AI controls
- AI shopping carts could drive up your grocery bill
- MOAE: Multi-Objective Agent Evolution with Pareto-Preserving Search
- Kazakhstan Tests AI After Finding 700 Errors in School Textbooks - The Times Of Central Asia
- I Let an AI Agent Hack All My Gadgets—and I’d Do It Again
- Cyderes CEO: ‘Not One’ CISO Has Disclosed Getting More Budget To Secure AI
Comments
Please log in to post a comment.