Mid-Tier AI Models Pose Growing Cybersecurity Threats, Researchers Warn

Researchers have raised concerns about the growing hacking capabilities of mid-tier AI models, which are described as cheaper and more efficient alternatives to advanced 'frontier' models. These mid-tier AI systems are improving rapidly in their ability to conduct cyberattacks, though specific technical details or metrics on their performance were not provided. The focus on these models contrasts with the current attention on high-profile AI threats, such as sandbox-escaping exploits in leading systems like those from Anthropic or OpenAI. No specific dates, vulnerabilities (e.g., CVE IDs), or organizations directly impacted were mentioned in the reported findings. The discussion stems from research highlighted in a CyberScoop article, emphasizing a shift in threat perception within AI-driven cybersecurity risks.