Anthropic makes Claude Code Auto Mode the default from August 14
Starting August 14, Claude Code Pro, Max and Team plans will run in Auto Mode by default. Anthropic says its classifier caught 89 percent of…
Read more
Desk archive
Tag: prompt injection
OpenAI’s GPT-Red Outperforms Humans in Automated Prompt Injection Red-Teaming
OpenAI has unveiled GPT-Red, an internal automated red-teaming model that significantly surpasses human red-teamers in identifying prompt injection vulnerabilities, demonstrating a new approach to AI…
Read more
OpenAI Uses AI to Enhance Model Security, Outperforming Human Red Teams
OpenAI has developed an internal AI model, GPT-Red, that significantly outperforms human red teamers in identifying security vulnerabilities in its AI systems, signaling a new…
Read more
WebMCP Tools Pose New AI Agent Hijacking Risks, Chrome Issues Security Guidance
New guidance from Chrome highlights how WebMCP-exposed tools can be exploited for AI agent hijacking through malicious manifests or contaminated outputs, shifting security responsibility to…
Read more
US Government Demands “Unhackable” LLMs from Anthropic, Raising Industry Concerns
The US government's insistence on "unhackable" large language models (LLMs) from developers like Anthropic, following the release of Fable 5, sparks debate on AI security…
Read more
OpenAI Introduces ‘Lockdown Mode’ for ChatGPT Business to Combat Prompt Injection Attacks
OpenAI has rolled out 'Lockdown Mode' for ChatGPT Business accounts and eligible personal users, aiming to enhance protection against sophisticated prompt injection attacks, particularly for…
Read more