Cheap Offense: A $3,000 Test Reached Inside OpenAI, and a Model Release Is Why

Cheap Offense: A $3,000 Test Reached Inside OpenAI, and a Model Release Is Why

0 View

Publish Date:
26 September, 2026
Category:
MSNBC
Video License
Standard License
Imported From:
Youtube

By Evan Vega

Three researchers at Hacktron AI ran an authorized bug-bounty test and reached OpenAI’s internal systems; OpenAI paid $6,500 and fixed it in about 14 hours. The story is not the hack. The whole two-month project cost under $3,000 in AI tokens, its hardest step succeeded only once Claude Opus 5 shipped mid-project, and the model refused the task until it was reworded. Meanwhile the US civilian cyber-defense agency, CISA, has lost about a third of its staff. Offense got cheap while public defense got cut. Reported at the journalism layer, no method described.


Read the full investigation →

Related: Frontier Watch

Read the full analysis →