Security researchers tricked LLMs into giving them cocaine recipes by abusing role models for prompt injection

The authors developed an attack called CoT (Chain of Thought) Forgery that involves using an LLM to spoof the terse style of ...

Prompt injection is exploiting enterprise AI's biggest design flaws by targeting agents, RAG pipelines and model routers

In 2025 and 2026, several independent sources have highlighted the same trend: Prompt injection remains one of the most ...
InfoQ

Prompt Injection for Large Language Models

Your LLM-based systems are at risk of being attacked to access business data, gain personal advantage, or exploit tools to the same ends. Everything you put in the system prompt is public data.
i-SCOOP

Token minimizing, how to cut LLM costs without losing quality

Token minimizing is the fastest way to lower LLM costs and latency. Learn practical techniques: prompt trimming, compaction, ...
Redmond Magazine

Microsoft Warns Harmful Prompt Attacks Can Undermine LLM Safety Controls

Microsoft research shows prompt-based attacks can bypass LLM safety guardrails and extract restricted information. GRPO safety training can be reversed via GRP-Obliteration using a single malicious ...
Fast Company

Why Anthropic’s decision to share the rules behind its Claude 3 chatbot is a big deal—sort of

The latest step forward in the development of large language models (LLMs) took place earlier this week, with the release of a new version of Claude, the LLM developed by AI company Anthropic—whose ...

When the Model Is Confident and Wrong: A Practitioner Guide to LLM Output Reliability

The model learns that hedging is a signal of lower-quality output. This creates a systematic bias toward sounding certain.