AI's 'Middle Class' Has Gotten Dramatically Better at Hacking

CyberScoop reports that mid-tier AI models have become dramatically more capable at offensive hacking, closing the gap with frontier models and broadening the pool of actors who can leverage AI for attacks. The finding extends the week's AI-weaponisation thread, following the disclosed near-autonomous AI attack on a Taiwanese government target.

Summary

The research indicates that mid-tier models now approach frontier-model offensive capability, lowering the barrier to effective cyber exploitation for a much wider range of actors. This sustains concerns about AI-accelerated vulnerability discovery and automation becoming a mainstream threat vector.

Impact

• Lowered barrier to entry for effective offensive hacking
• Broader pool of actors capable of AI-enabled exploitation
• Sustains the AI-weaponisation threat theme heading into coming weeks

Sector

Global (Macro)

Sources

• CyberScoop — Mid-Tier AI Models and the Hacking Threat
• raw/digests/Cyber-Digest-2026-08-14