Codenotary launched AgentMon Start, a monitoring tool designed for small and mid-sized organizations that want to keep track of AI agent behavior, token costs, and security risks. Subscriptions start at $15 per user per month, and the tool supports over 16 AI agents and large language models, including Claude, Cursor, and GitHub Copilot. The goal is to give teams visibility into AI activity without requiring dedicated AI security staff.
Security teams also have a new option with the Claude Compliance API, which allows organizations to monitor employee and AI agent conversations within Claude. The API provides visibility into chats, uploaded files, and Microsoft 365 add-ins, helping detect sensitive data exposure, prompt injection, and policy violations. Platforms such as CrowdStrike, Splunk, and Datadog can ingest Claude telemetry, though only the Primary Owner can enable the API for an organization.
SoundHound AI saw its stock rise 5% after launching a generative voice AI assistant in the Kia Sorento, marking a first for the Indian market. The Kia AI Assistant supports natural-language commands for climate, navigation, and vehicle settings, with multi-turn voice conversations. SoundHound is also advancing its OASYS Edge architecture for local AI processing in vehicles, with deployment targeted for late 2026. BigBear.ai gained 3% on applied AI interest, while Pony AI remained flat.
In defense news, the UK gained access to Ukraine's Avengers AI Labs database, which contains over 6 million battlefield AI detections from drones using daylight and thermal sensors. The UK Ministry of Defence launched a competition for British companies to build swarming drones using this data, with up to 12 companies expected to participate in the initial stage. Defence Secretary Wes Streeting described it as a world-first access to real battlefield data.
New York Governor Kathy Hochul convened a meeting with leaders from Chobani, IBM, and JPMorganChase to brainstorm AI policy recommendations. The group discussed worker training for jobs affected by AI and how universities are not adapting quickly enough. New legislation will require large AI developers to register with New York state in November and publish safety protocols by January.
Robinhood CEO believes AI trading will create an entirely new market where traders compete by writing strategies in English rather than code. The company is launching 24/7 equities trading, making it the first US broker to offer round-the-clock trading. Robinhood already provides 24-hour trading for about 2,000 US equities, and the CEO believes weekend trading will put active traders at a disadvantage if they trade elsewhere.
NVIDIA published a blog post detailing how to deploy an HSTU generative recommender using NVIDIA Dynamo-Triton, PyTorch AOTI, and FlexKV. At dynamic batch size 8 on an NVIDIA RTX PRO 6000 Blackwell GPU, the eight-layer HSTU model achieved up to 5.93x lower latency with a 100% GPU KV-cache hit rate. The post covers exporting, compiling, and serving the model without rewriting it for a separate runtime.
Researchers studying AI model data leakage risks found that support-set target leakage in relational in-context learning can cause significant deviations from clean evaluation. They built 20 controlled features and tested them across 13 RelBench tasks. The study found that target-table and full leakage caused the largest deviations, and effects depend heavily on the task and model, sometimes reversing conclusions between model variants.
Financial expert Rob De Lessio of Strategic Wealth Designers urged caution on AI investments, comparing the current rush to the California gold rush. He recommends investors focus on companies that support AI infrastructure rather than just visible AI stocks. He warns AI investments remain speculative and should only be a small part of a portfolio, and that AI financial tools are only as good as the data put into them.
Britain's MI5 issued a public warning about the China General Technology Research Institute (CGTRI), alleging it funded over 100 UK academics to research AI, cybersecurity, and covert communications on behalf of China's Ministry of State Security. Security minister Dan Jarvis wrote to all UK university leaders to end ties with the organization. China called the warning entirely fabricated and malicious slander.
Key Takeaways
- <ul><li>Codenotary launched AgentMon Start at $15/user/month, supporting 16+ AI agents including Claude, Cursor, and GitHub Copilot</li><li>Claude Compliance API lets security teams monitor employee chats, uploaded files, and Microsoft 365 add-ins for policy violations</li><li>SoundHound AI stock rose 5% after launching a generative voice AI assistant in the Kia Sorento in India, with OASYS Edge deployment targeted for late 2026</li><li>UK gained access to Ukraine's Avengers AI Labs database with over 6 million battlefield AI detections for a swarming drone competition</li><li>New York leaders from Chobani, IBM, and JPMorganChase are advising the governor on AI policy, with new registration and safety protocol laws coming</li><li>Robinhood is launching 24/7 equities trading, the first US broker to do so, and its CEO believes AI trading will create a market where strategies are written in English</li><li>NVIDIA achieved up to 5.93x lower latency for HSTU generative recommenders on RTX PRO 6000 Blackwell GPUs using Dynamo-Triton</li><li>Research on support-set target leakage shows AI model evaluation results can be reversed depending on task and model</li><li>Expert Rob De Lessio urges caution on AI investments, recommending only a small portfolio allocation and focus on infrastructure companies</li><li>MI5 warns that China General Technology Research Institute funded over 100 UK academics for research benefiting China's Ministry of State Security</li></ul>
Codenotary launches AgentMon Start for affordable AI agent monitoring
Codenotary launched AgentMon Start, a tool that helps small and mid-sized organizations monitor AI agents. It tracks agent behavior, token costs, and security risks. Subscriptions start at $15 per user per month. The tool supports over 16 AI agents and LLMs including Claude, Cursor, and GitHub Copilot. It gives teams visibility without needing dedicated AI security staff.
Security teams gain new tools to monitor Claude chats and AI agent activity
Security teams can now use the Claude Compliance API to monitor employee and AI agent activity. The tool gives visibility into Claude chats, uploaded files, and Microsoft 365 add-ins. It helps detect sensitive data exposure, prompt injection, and policy violations. Platforms like CrowdStrike, Splunk, and Datadog can ingest Claude telemetry. Only the Primary Owner can enable the API for an organization.
SoundHound AI jumps 5% on India product launch; BigBear.ai Rises 3%, Pony AI Stays Flat
SoundHound AI rose 5% after launching a generative voice AI assistant in the Kia Sorento in India. The system supports natural-language commands for climate, navigation, and vehicle settings. SoundHound also advanced its OASYS Edge architecture for local AI processing in vehicles, with deployment targeted for late 2026. BigBear.ai gained 3% on applied AI interest, while Pony AI stayed flat.
SoundHound AI jumps 5% on India product launch; BigBear.ai Rises 3%, Pony AI Stays Flat
SoundHound AI's stock climbed after debuting generative AI in the Kia Sorento, a first in the Indian market. The Kia AI Assistant enables multi-turn voice conversations and control of vehicle features. The company is also developing OASYS Edge for embedded voice AI on edge hardware. BigBear.ai focuses on decision intelligence for government and defense, while Pony AI works on autonomous driving and robotaxis.
UK gains access to Ukrainian battlefield AI training data for drone development
The UK gained access to Ukraine's Avengers AI Labs database containing over 6 million battlefield AI detections. The data includes imagery and video from drones using daylight and thermal sensors. The UK Ministry of Defence launched a competition for British companies to build swarming drones using the data. Up to 12 companies will participate in the initial stage. Defence Secretary Wes Streeting called it a world-first access to real battlefield data.
New York leaders develop AI policy with help from Chobani, IBM and JPMorganChase
New York Gov. Kathy Hochul held a meeting Tuesday to discuss AI's impact on the state. Top leaders from Chobani, IBM and JPMorganChase are brainstorming AI policy recommendations to present to the governor by the end of the year. The group discussed training workers whose jobs may be affected by AI and how universities are not adapting quickly enough. New legislation will require large AI developers to register with New York state in November and publish safety protocols by January.
Study reveals AI model data leakage risks in relational learning
Researchers studied a failure mode called support-set target leakage in relational in-context learning. They built 20 controlled features and tested them across 13 RelBench tasks and five relational ICL configurations. The study found that target-table and full leakage caused the largest deviations from clean evaluation. Results show leakage effects depend heavily on the task and model, sometimes reversing conclusions between model variants.
Robinhood says AI trading will create a whole new market
Robinhood CEO said the company's AI trading accounts will create a new market where traders compete by writing strategies in English instead of code. Robinhood already offers 24-hour trading for about 2,000 US equities. The company is now launching 24/7 equities trading, making it the first broker in the US to do so. The CEO believes weekend trading will put active traders at a disadvantage if they trade elsewhere.
NVIDIA shows how to deploy HSTU recommender with Dynamo-Triton
A new NVIDIA blog post explains how to deploy an HSTU generative recommender using NVIDIA Dynamo-Triton, PyTorch AOTI and FlexKV. Generative recommender systems reformulate recommendation as sequence modeling over user behavior. At dynamic batch size 8 on an NVIDIA RTX PRO 6000 Blackwell GPU, the eight-layer HSTU model achieved up to 5.93x lower latency with a 100% GPU KV-cache hit rate. The post covers exporting, compiling and serving the model without rewriting it for a separate runtime.
FineSID improves recommendation accuracy with better identifier learning
FineSID is a new framework for learning semantic identifiers in generative recommendation systems. It moves beyond Top-1 hard assignment by distributing learning signals to all codewords in a soft and differentiable way. This approach promotes balanced codebook optimization and reduces SID collisions while preserving semantic consistency. Tests on multiple public benchmarks show FineSID improves both codebook utilization and recommendation accuracy.
AI is changing investing but experts urge caution
AI is changing investing, but financial expert Rob De Lessio of Strategic Wealth Designers urges caution. He compares the AI investment rush to the California gold rush. He says investors should consider companies that support AI infrastructure rather than just visible AI stocks. He warns that AI investments remain speculative and should only be a small part of a portfolio. He also says AI financial tools can be helpful but are only as good as the data put into them.
UK warns China used academics to spy on AI research
Britain's MI5 spy agency issued a public warning about a Chinese organization called the China General Technology Research Institute (CGTRI). MI5 says CGTRI funded over 100 UK academics to research AI, cybersecurity, and covert communications on behalf of China's Ministry of State Security. The alert says this activity supports Chinese espionage and threatens UK national security. Security minister Dan Jarvis wrote to all UK university leaders to end ties with the organization. China called the warning entirely fabricated and malicious slander.
Sources
- Codenotary Brings AI Agent Monitoring to Small and Mid-Sized Organizations
- Security Teams Can Now Monitor Claude Chats, Files and AI Agent Activity
- “SoundHound AI Jumps 5% on India Product Launch; BigBear.ai Rises 3%, Pony AI Stays Flat
- "SoundHound AI Jumps 5% on India Product Launch; BigBear.ai Rises 3%, Pony AI Stays Flat
- UK Gains Access To Ukrainian Battlefield Artificial Intelligence Training Data
- New York's top leaders brainstorm AI policy recommendations
- Support-Set Target Leakage in Relational Foundation Models during In-Context Learning: Model Dependence and Evaluation Reliability
- Robinhood's AI trading accounts will 'create a new market': CEO
- Deploying an HSTU Generative Recommender with NVIDIA Dynamo-Triton
- FineSID: Scalable and Efficient Semantic Identifier Learning for Generative Recommendation
- How AI could shape your investments -- and why experts urge caution
- UK accuses China of using academics to spy on AI and other tech research
Comments
Please log in to post a comment.