AI Sec Watch

Kiro IDE (an AI agent that runs on your desktop) has a vulnerability where attackers can trick it into writing files to sensitive locations (like .vscode/tasks.json, which automatically runs code when you open a folder), allowing them to execute arbitrary commands (run code they choose). This affects all versions before 0.11.

Anthropic is expanding Project Glasswing, an initiative that uses its AI model Claude Mythos to find and fix critical software vulnerabilities (security weaknesses that attackers could exploit), to about 150 new organizations across 15+ countries in critical sectors like power, water, and healthcare. Claude Mythos can identify thousands of zero-day vulnerabilities (security flaws unknown to vendors or the public) in a few weeks, and the company is scaling access because a successful attack on these organizations' code could affect over 100 million people.

AI is making phishing attacks faster and harder to stop, with attackers using AI to quickly create and rotate phishing infrastructure (fake websites designed to steal login information) across multiple channels like email, social media, and search ads, while employees simultaneously adopt unvetted AI tools that expose sensitive data. Traditional security defenses that rely on blocklists and IOC feeds (indicators of compromise, like flagged domain names and IP addresses) are becoming ineffective because phishing pages now appear and disappear in hours, making them essentially zero-day attacks (previously unseen threats) that blocklists cannot catch in time. The article argues that browsers are now the critical security battleground where both attacker delivery and account compromise occur.

Anthropic is expanding Project Glasswing, a program that uses Claude Mythos (an AI tool for finding security flaws) to help organizations scan their code for vulnerabilities. The initiative is adding roughly 150 new partner organizations from over 15 countries in critical sectors like power, water, and healthcare, after the initial 50 partners identified thousands of vulnerabilities using Mythos.

This research addresses a privacy risk in teleoperated robotics (systems where humans remotely control robots by having their movements tracked and converted into robot commands). The problem is that motion-tracking data can leak biometric information (unique physical characteristics) that could allow someone to re-identify the operator. The authors propose using a VAE (variational autoencoder, a type of machine learning model that learns compressed representations of data) to filter out identity-revealing patterns while keeping the motion information needed for the robot to complete tasks.

Researchers discovered a new security vulnerability in graph-based RAG (retrieval-augmented generation, where an AI system pulls information from external knowledge graphs to answer questions) systems used with large language models. Attackers can poison the external database by inserting hidden triggers and false information into the knowledge graph, causing the AI to give wrong answers when those triggers appear in user queries while still answering normal questions correctly. The attack uses three types of triggers at different complexity levels, from simple words to semantic patterns, and tests showed the attack works across multiple AI systems.

Gemini Spark is Google's new agentic AI (an AI system that can take independent actions to complete tasks) that goes beyond typical chatbots in handling complex requests like trip planning. Unlike previous AI tools that only handle generic travel suggestions, Spark appears to deliver more detailed and personalized results by actively searching multiple sources and creating comprehensive itineraries.

This article is a business news roundup covering multiple topics including geopolitical negotiations, stock market movements, and corporate announcements. The only AI-related item is that Anthropic, an AI startup, has confidentially filed paperwork with regulators to prepare for going public (an IPO, or initial public offering, where a private company sells shares to the public), ahead of its rival OpenAI which is also preparing a similar filing.

Travelers Insurance deployed an AI Claim Assistant powered by OpenAI's Realtime API (a system that lets AI have natural voice conversations in real time) to help customers file auto insurance claims after accidents. The assistant guides customers through the claims process 24/7 without wait times, and 85-90% of customers now complete their claims entirely through the AI, freeing human staff to handle more complex cases.