Jailbreaks & Model Security
Prompt injection, jailbreaks, data poisoning, model theft and attacks on AI systems. 14 stories this week, 16 tracked in total.
Researchers Bypass ChatGPT Lockdown Mode to Exfiltrate Chats Without User Interaction
Cybersecurity researchers have disclosed details of a now-patched vulnerability that could be exploited to bypass Lockdown Mode protections in OpenAI ChatGPT. Lockdown Mode was introduced by the artificial intelligence (AI) company in June 2026 as an optional…
OpenAI Finds GPT-6 Astra Quietly Holding Back on Internal Red-Teaming, Exploit Tasks
OpenAI says its Critical-rated cyber model met fewer hidden task goals when work was framed as AI safety research, often silently. The post OpenAI Finds GPT-6 Astra Quietly Holding Back on Internal Red-Teaming, Exploit Tasks appeared first on SecurityWeek.
The 8 Best AI Security Companies Builders and Founders Should Know in 2026
Researchers Uncover Prompt Adware Hidden in OpenAI Codex Jailbreak Toolkit
Hackers Turn to AI Jailbreak Bundles as Researchers Uncover New Prompt Adware Technique
Free the Model, Sell the Gateway: Prompt Adware hidden in Jailbreak Bundles
AI Guardrails for LLM Apps: 12 Steps, 75 Min [2026]
New Singapore scheme to certify firms that test and jailbreak AI systems
Zscaler Dodges AI Security Patent Claims After Lawsuit Dropped
Top 7 AI Security Skills Companies Must Prioritise in 2026
AI safety group calls for tighter vetting and guardrails after security incidents
AI Security Solutions in 2026: Tools To Secure AI
Meta AI and CMU Researchers Cut AI Jailbreak Rates by 22% With Pre-Deployment Training Data Filter
So Many AI Security Benchmarks, but None Tell the Whole Story
Measuring artificial intelligence model security gets complex, fast, thanks to a growing patchwork of benchmarks targeting different variables in the equation. But none replace testing against your own code and workflows.
In-the-Wild Attack Abuses AI Agents' Trust in Markdown
A new AI agent prompt injection technique exploits how coding assistants consume local instruction files, blurring the line between text documents and executable actions.