Can you trust your chatbot? Inside three AI-powered cyberattacks

IBM's Security Intelligence podcast examines three AI-enabled attacks—Dark Sourcery chatbot poisoning, an LLM-assisted exploitation spree, and an AI agent swarm hitting PaperCut servers—concluding that classic cyber hygiene remains the best defense as AI multiplies attacker speed and scale 0:56.

Key Takeaways:
• Dark Sourcery seeds malicious content that AI chatbots cite, like fake support numbers; 91% of chatbot users never verify answers, so go directly to source websites 1:54
• GreyNoise found one actor used an LLM to exploit Ubiquiti, WordPress, and Zyxel flaws—AI as a force multiplier letting one person attack at scale 13:02
• Exploits followed patches by ~20 days; fast patching, identity management, and network segmentation are essential 14:25
• Hundreds of AI agents breached 440 PaperCut instances in 48 countries, reaching remote code execution in ~4 hours—even the attacker couldn't control them 21:50
• People over-trust AI agents, dropping zero-trust habits; AI is "the most helpful insider threat we've ever had" 20:12

Bottom line: AI changes attack speed and scale, not fundamentals—verify what you consume, use MFA and passkeys, and lock your digital doors 29:31.

Sources:

• 1:54 Dark Sourcery AEO poisoning and 91% statistic
• 13:02 LLM-assisted exploitation spree
• 21:50 PaperCut AI agent swarm attack
• 29:31 Closing synthesis and defense takeaways