Researchers have discovered that a prompt can be disguised as an url, and accepted by Atlas as an url in the omnibox.
The post ChatGPT Atlas’ Omnibox Is Vulnerable to Jailbreaks appeared first on SecurityWeek.
UAE’s K2 Think AI Jailbroken Through Its Own Transparency Features
Researchers exploited K2 Think’s built-in explainability to dismantle its safety guardrails, raising new questions about whether transparency and security in AI can truly coexist.
The post UAE’s K2 Think AI Jailbroken Through Its Own Transparency Features appeared first on SecurityWeek.
GPT-5 Has a Vulnerability: Its Router Can Send You to Older, Less Safe Models
Instead of GPT-5 Pro, your query could be quietly redirected to an older, weaker model, opening the door to jailbreaks, hallucinations, and unsafe outputs.
The post GPT-5 Has a Vulnerability: Its Router Can Send You to Older, Less Safe Models appeared first on SecurityWeek.
Red Teams Jailbreak GPT-5 With Ease, Warn It’s ‘Nearly Unusable’ for Enterprise
Researchers demonstrate how multi-turn “storytelling” attacks bypass prompt-level filters, exposing systemic weaknesses in GPT-5’s defenses.
The post Red Teams Jailbreak GPT-5 With Ease, Warn It’s ‘Nearly Unusable’ for Enterprise appeared first on SecurityWeek.
AI Guardrails Under Fire: Cisco’s Jailbreak Demo Exposes AI Weak Points
Cisco’s latest jailbreak method reveals just how easily sensitive data can be extracted from chatbots trained on proprietary or copyrighted content.
The post AI Guardrails Under Fire: Cisco’s Jailbreak Demo Exposes AI Weak Points appeared first on SecurityWeek.
Grok-4 Falls to a Jailbreak Two days After Its Release
The latest release of the xAI LLM, Grok-4, has already fallen to a sophisticated jailbreak.
The post Grok-4 Falls to a Jailbreak Two days After Its Release appeared first on SecurityWeek.
New AI Jailbreak Bypasses Guardrails With Ease
New “Echo Chamber” attack bypasses advanced LLM safeguards by subtly manipulating conversational context, proving highly effective across leading AI models.
The post New AI Jailbreak Bypasses Guardrails With Ease appeared first on SecurityWeek.
New Jailbreak Technique Uses Fictional World to Manipulate AI
Cato Networks discovers a new LLM jailbreak technique that relies on creating a fictional world to bypass a model’s security controls.
The post New Jailbreak Technique Uses Fictional World to Manipulate AI appeared first on SecurityWeek.
New CCA Jailbreak Method Works Against Most AI Models
Two Microsoft researchers have devised a new jailbreak method that bypasses the safety mechanisms of most AI systems.
The post New CCA Jailbreak Method Works Against Most AI Models appeared first on SecurityWeek.
DeepSeek Compared to ChatGPT, Gemini in AI Jailbreak Test
DeepSeek’s susceptibility to jailbreaks has been compared by Cisco to other popular AI models, including from Meta, OpenAI and Google.
The post DeepSeek Compared to ChatGPT, Gemini in AI Jailbreak Test appeared first on SecurityWeek.