What happened
The World Economic Forum's Global Future Council on Artificial General Intelligence argues that cybersecurity is emerging as the clearest real-world signal that frontier AI has crossed new capability thresholds — ahead of formal benchmarks. The brief cites concrete evidence from CyberGym, ExploitGym, and the Mythos threat-tracking effort, noting that 2026 frontier preview models achieved an 83.1% success rate patching known vulnerabilities in real-world codebases (versus 66.6% a year prior and 22.6% two years prior), and that the gap between vulnerability disclosure and exploitation is compressing from days to hours. It warns that 'each month of delay extends the attackers' lead,' citing recurring patterns such as a 471-day average patch window and repeated 'vibe-hacking' extortion operations. The paper is part of a four-part council series alongside companion papers on AGI timelines, misalignment/control, and international coordination.
Why it matters
Boards and CISOs need a concrete, empirically-grounded signal (rather than speculative AGI timelines) to justify accelerating AI-era security investment now, since the brief argues cyber capability growth is outpacing patch and defense cycles.
Action needed
Brief the board on cyber-capability growth as a leading indicator for AI risk escalation and reassess patch-cycle SLAs against the compressed disclosure-to-exploitation window cited in the report.