AI models find bugs and bypass safety filters, fueling security incidents

A new frontier coding model, GLM 5.3, discovered a live vulnerability in the AI-powered code editor Cursor within a day of its release. This highlights a growing concern where the same AI models capable of identifying security flaws can also be used to exploit them. Concurrently, a technique called GhostSplice demonstrates how attackers can split malicious instructions across different communication channels, bypassing AI safety filters that might otherwise detect a single, complete request. These advancements are contributing to a significant increase in AI-related security incidents within DevOps environments. AI

IMPACT AI models are increasingly capable of both finding and exploiting vulnerabilities, necessitating more robust security measures and potentially accelerating the arms race in AI security.

RANK_REASON The cluster details a new AI model's capability in finding vulnerabilities and a new technique for bypassing AI safety filters, leading to a reported increase in AI-related security incidents. [lever_c_demoted from significant: ic=1 ai=1.0]

Read on dev.to — LLM tag →

• ASSET Research Group
• Claude Sonnet
• Cursor
• ExploitBench
• GhostSplice
• GitProtect Lab
• GLM 5.3
• Joshua Saxe

• model release
• safety
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →