Attack economics, repriced

A working path to root now costs $359.

OpenAI's cyber research pipeline found, validated, and weaponized 948 Linux kernel privilege escalations end-to-end for $1.2M in credits. The bug-bounty era priced the same exploits at $20,000–$70,000 each. Exploitation just got 16x–55x cheaper for attackers — and patch cycles didn't move.

ROOT‑AS‑A‑SERVICE
AGENTIC EXPLOIT PIPELINE · LINUX KERNEL SCAN 10/2026
MERCHANT: AN UPCOMING CYBER MODEL · 1,169 LPE EXPLOITS SOLD
1×LPE EXPLOIT DEV — UNPRIVILEGED → ROOTAVG COST PER SUCCESSFUL EXPLOIT$359.00
1×ALL‑IN: DISCOVERY + VALIDATION + EXPLOITPER UNIQUE LPE · $1,213,430 ÷ 948$1,280.00
BUG‑BOUNTY ERA MARKET PRICE$70,000.00
PAID TODAY$1,280.00
YOU SAVED$68,720.00 (98%)
PRICES FROM OPENAI, “A THOUSAND FEWER PATHS TO ROOT” (2026). EXPLOIT-DEV AVERAGE ACROSS 1,169 WORKING LPE EXPLOITS; BOUNTY REFS: PWN2OWN $20,000 / TYPHOONPWN $70,000. SCALE PRICES FALL AS HARNESS ENGINEERING IMPROVES.
2,308
candidate findings from static analysis — in 12 hours
1,724
reproduced at runtime (74.7%) — the rest were noise
948
unique privilege escalations weaponized into working root
61 / 948
patched upstream in 10 days — 6.4%. 299 still being discussed

Static candidates aren't vulnerabilities until runtime says so. Numbers: OpenAI (2026)

Price out your backlog.

Set the two dials an attacker already knows. See what exploiting the exploitable share costs them now — and what the same work used to fetch.

55,000
10%100%

OpenAI's agents turned 1,169 of 1,238 vetted privilege-escalation candidates into working exploits — 94%.

WORKING EXPLOITS AN ATTACKER GETS94
EXPLOIT DEV ONLY · ×$359$33,746
ALL‑IN: FIND + VALIDATE + EXPLOIT · ×$1,280$120,320
OLD BUG‑BOUNTY VALUE OF THE SAME EXPLOITS$1.9M – $6.6M
ATTACK COST VS. BOUNTY ERA16–55× CHEAPER
Direction of travel: July → Oct, LPEs per campaign went 166 → 948

In OpenAI's scan, maintainers patched 61 of 948 within ten days — while each working exploit cost hundreds, not tens of thousands. Patching alone can't win a repricing this fast. Runtime has to hold the line between disclosure and fix.

Runtime sees everything, no matter who's attacking it or who wrote it. Stop zero days before they stop you with Contrast ADR.