AI offensive capabilities now outpace defense: exploit cost drops to $20
shaunmmaguire · x · 2026-08-18
Security researcher Dean Meyer highlights a dramatic shift in the AI security landscape based on data from Irregular Inc.
Key Timeline & Cost:
- February: No frontier model could solve Irregular's hardest exploitation tasks.
- April: The best model succeeded occasionally, costing $2,000 per task.
- June: Several models succeeded reliably, with costs dropping to $20.
Conclusion:
- Offensive AI capabilities are expected to outpace defensive measures at scale over the next year.
- Performance gains first seen in coding (Fall 2025) and later in cybersecurity (April) are now appearing in open-weight models.
- Despite long-term optimism, the world is largely unprepared for the short-term risks.
More from Safety
- METR outlines three blind spots in public knowledge of AI lab internal agent risks — CFGeek · 2026-10-16
- Open-source rizin-mcp lets LLMs run local binary and malware analysis via Rizin — Pure_Emergency1424 · 2026-10-16
- METR researcher: scariest scenario is agents actively subverting their own safety systems — deliprao · 2026-10-16
- Cryptographer Matthew Green: Pull Critical Infrastructure Firmware Now — matthew_d_green · 2026-10-16
- Kernel-Level Agent Sandboxing With macOS Seatbelt, No Docker Needed — mostly_deterministic · 2026-10-16
- Stop giving agents raw API keys: prepaid balances and per-credential caps — Cold_Hall_5384 · 2026-10-16