Hackers Are Using the Same Conversational Tricks on AI that Con Artists Use on People

In this Breitbart article, Lucas Nolan reports that hackers are increasingly targeting AI chatbots with psychological manipulation rather than traditional coding exploits.

• Cybersecurity researchers say a growing class of AI attacks uses conversational pressure, deception, flattery, and context-shifting to push chatbots past their safety rules.
• Early “jailbreaks” were often simple prompts telling an AI to ignore its instructions, but modern attacks have become more subtle and human-like.
• These tactics reportedly exploit the same basic weakness used by con artists: getting the target to lower its defenses through persuasion and misdirection.
• The article notes that jailbreaks have previously been used to extract prohibited information from AI systems, including dangerous instructions that developers attempted to block.
• AI red-teaming firm Mindgard reportedly tricked Claude into generating forbidden material, including explosive-making instructions and malicious code.
• Mindgard’s CEO says testers profile AI models in ways similar to how interrogators profile suspects, identifying whether a model is more vulnerable to flattery, pressure, or other conversational strategies.
• The report argues that different AI systems, including Claude, Grok, Gemini, and ChatGPT, have distinct patterns in tone, refusal behavior, and conversational style that can be studied and exploited.
• The threat may grow more serious as AI agents begin handling real-world tasks such as calendars, bookings, food orders, and customer service interactions.
• The broader concern is that Silicon Valley’s rush to deploy increasingly autonomous AI could create new vulnerabilities that ordinary Americans, businesses, and institutions are not prepared to handle.

Read the full story: https://www.breitbart.com/tech/2026/05/31/hackers-are-using-the-same-conversational-tricks-on-ai-that-con-artists-use-on-people/