AI Agents in the Crosshairs: Attacks on Stores and New Research

In today’s roundup: a report on online store hacks using AI agents and research on making models and agentic systems more reliable.

🔥 Hot:

🔹 Gambit Security reports AI-agent attacks on online stores — The blog post’s headline says that a company could become the target of an attack for $25.

➡️ News:

🔹 Lightspeed targets a $250 million fund for early-stage AI startups in India — For the first time, the company is synchronizing its Indian fundraising cycle with its global funds. 🔹 A technical report introduces the Pistis family of multimodal models — It includes 27B- and 9B-parameter models built on Qwen3.6 and Qwen3.5.

➡️ Useful materials:

🔹 DEEPO proposes improving multimodal model training to combat hallucinations — The authors examine where weaknesses arise in the chain from reward to parameter updates. 🔹 BaseCamp uses an AI agent to configure DNA sequencing pipelines — The system helps make decisions about quality thresholds and other settings that usually require manual work. 🔹 Researchers examined when forecasting agents should search for data and reason — The paper compares these actions with relying on a market forecast or historical analogies. 🔹 Habr examines the legal issues AI faces in Russia — The article is based on a talk for law students.

📝 If you would like to add other news and materials to the list, write in the comments.