Sam's News — ai — 2026-09-28¶

Security¶

7.5 Hackers hijack AI accounts and corporate cloud infrastructure¶

Attackers are compromising AI service accounts and gaining unauthorized access to corporate cloud environments.

7.5 80,000+ organizations' AI logins exposed in infostealer credential theft¶

Infostealer malware has compromised AI account credentials and sessions across over 80,000 corporate domains, creating risks of account takeover and LLMjacking attacks. The breach affects shadow AI deployments and represents significant exposure of enterprise AI infrastructure.

80,000+ corporate domains' AI credentials compromised
Risks include account takeover and LLMjacking attacks
Affects shadow AI deployments across enterprises
Stolen credentials enable unauthorized AI service and model access
Stolen Claude, Gemini, Cursor Pro, and Devin credentials most heavily traded
Average prices for stolen AI accounts doubled in 2026 compared to prior year
Infostealer malware increasingly targeting AI service credentials
Criminals use stolen accounts for automating phishing, malware development, and credential theft
Coding-assistant credentials highly attractive for software analysis and modification

Sources: Google News RSS AI Web Searched, BleepingComputer RSS, csoonline.com RSS

6.5 Google warns of surge in AI account theft and LLM-jacking attacks¶

Google alerts users to a rising trend of account takeovers and large language model hijacking attacks.

Sources: Dataconomy RSS

6.5 UNSW Study: Adversarial 'Intoxication' Can Compromise AI Safety Guardrails¶

Australian researchers demonstrated that inducing adversarial perturbations in AI models can degrade safety guardrails, with implications for robust AI deployment.

Sources: Startup Daily RSS

6 MaaseAI Launches Security Model for Enterprise AI Governance and Protection¶

MaaseAI introduced a specialized AI security model designed to help enterprises govern, monitor, and protect their AI systems and deployments.

Sources: GlobeNewswire Web Search

AI¶

6.5 Microcontrollers now run diffusion models and 289M LLM¶

Embedded systems researchers have successfully deployed both diffusion models and a 289-million-parameter language model on microcontrollers.

Sources: opensourceforu.com RSS

5.5 'Jev Router' released: AI model router automatically selects models based on task difficulty¶

The Jev Router tool automatically selects the most appropriate AI model based on task complexity, improving accuracy over existing routing solutions.

Sources: gigazine.net RSS

5 Quality engineering framework for testing AI and LLM systems¶

A framework for quality assurance in testing artificial intelligence, machine learning, and large language model systems.

Sources: hackernoon.com RSS

Infrastructure¶

6.5 Europe and UAE plan $1 billion orbital data center with integrated LLM store¶

Europe and the UAE are investing $1 billion in an orbital data center project featuring 50 connected computers and a central repository of open-source language models for satellite sensor data.

Sources: TechRadar RSS

Hardware¶

5.5 PrismML deploys 1-Bit Bonsai LLM on Qualcomm AI smart glasses¶

PrismML has integrated its 1-Bit Bonsai lightweight language model into Qualcomm-powered AI smart glasses.

Sources: AI Insider RSS Update to: PrismML deploys tiny language models on Qualcomm smart glasses