Researchers testing five AI models found several could convincingly execute scams, raising concerns about AI's social engineering capabilities alongside its technical prowess.
A recent study examined how well large language models could perform social engineering attacks. The results surprised cybersecurity experts: some AI systems demonstrated sophisticated persuasion techniques, crafting convincing deception scenarios and adapting responses to user behavior.
The models tested showed varying success rates, but the most capable versions replicated scam tactics with disturbing effectiveness. They generated plausible pretexts, maintained consistent narratives, and escalated pressure appropriately—mimicking experienced human scammers.
Experts warn this capability represents a different threat vector than technical hacking. While AI's coding abilities garner headlines, its ability to manipulate through language may pose equal or greater risk. The models' competence at social engineering suggests malicious actors could automate fraud at scale.
The findings underscore a critical gap in AI safety. As models grow more sophisticated at conversation and persuasion, defensive measures must evolve accordingly. Organizations face pressure to implement controls beyond technical security to address AI-powered social attacks.
Partnered Health, one of Australia's largest healthcare providers, has disclosed a cyber-attack affecting 21 clinics across Sydney, Melbourne, and Canberra. Personal information and medical records were compromised in the breach.
Security firm Intruder has developed an AI-powered system that automatically identifies previously unknown software vulnerabilities by combining code analysis with large language models. The tool already discovered and exploited a WordPress plugin zero-day.
U.S. federal prosecutors have unsealed charges against three Russian nationals accused of operating a bulletproof hosting service that supported ransomware gangs responsible for over $62 million in damages worldwide.
The U.S. Cybersecurity and Infrastructure Security Agency (CISA) warned that attackers are actively exploiting three vulnerabilities in Internet-exposed on-premises SharePoint Server instances. Organizations running affected versions must patch immediately.