#LLM
All CosmicBytez Labs articles tagged #LLM, across news, security advisories, how-to guides, and projects.
- News
Claude Opus 5.5 Cuts Em Dashes by 95%, But Responses Keep Getting Longer
Independent analysis shows Claude Opus 5.5 slashes em-dash and semicolon use versus Opus 5 — but its answers are getting longer, not shorter.
- News
AI Recommendation Poisoning: How "Ask AI" Buttons Silently Alter LLM Memory
A new class of prompt injection embedded in "Ask AI" buttons on commercial websites silently writes biased recommendations into users' AI assistant memory...
- News
We Built a Vulnerability Vending Machine: AI Tokens In, Zero-Days Out
Security firm Intruder built an AI-powered system that combines code slicing with large language models to automatically discover complex software...
- News
'Ghostcommit' Hides Prompt Injection in Images to Fool AI Agents and Steal Secrets
Security researchers have demonstrated 'Ghostcommit,' a technique that embeds prompt injection payloads inside PNG images to bypass AI code review tools...
- News
JadePuffer: The First Fully Autonomous LLM-Driven Ransomware Attack
Security researchers at Sysdig have documented JadePuffer — an agentic threat actor powered entirely by a large language model that independently...
- News
Chinese LLMs Broaden the Gap Between Attackers & Defenders
China's GLM 5.2 finds vulnerabilities at $0.17 each and outperforms frontier Western models on security benchmarks — while AI-enabled adversary activity...
- News
JadePuffer Ransomware Used an AI Agent to Automate the Entire Attack
Security researchers have documented what appears to be the first ransomware operation conducted entirely by a large language model agent — JadePuffer...
- News
Claude Fable 5 Isn't Permanently Leaving Subscriptions, Anthropic Says
Anthropic confirms Claude Fable 5 will leave Pro, Max, and Team subscription plans on July 7, shifting to usage credits priced at $10/M input and $50/M...
- News
Claude Fable Relaunch Disappoints Users With Nerfed Performance
Claude Fable 5's July 1 global relaunch revealed a new safety classifier that silently reroutes sessions involving security-related keywords to the weaker...
- News
Weekly Recap: Telecom Sleeper Cells, LLM Jailbreaks, Apple
This week's cybersecurity roundup covers long-running telecom espionage operations reaching courtrooms, resurging LLM jailbreak techniques, Apple's UK age...
- News
Alibaba Launches Qwen 3.5 — Claims to Outperform GPT-5.2
Alibaba Cloud unveils Qwen 3.5, a 397-billion-parameter Mixture-of-Experts AI model with native multimodal capabilities, claiming 60% lower costs and...