'Mayhem' Wins $2M In DARPA's AI Hacking Contest, Draws EFF Scrutiny

Here's the highlight reel from the DARPA-sponsored "Cyber Grand Challenge" competition. Slashdot reader alphadogg writes: Cyber-reasoning platform Mayhem pulled down the $2 million first prize in a competition...that pitted entrants against each other in the classic hacking game Capture the Flag, never before played by programs running on supercomputers. A team from Carnegie Mellon University spin-out All Secure entered Mayhem in the competition against six other programs played in front of thousands in the ballroom of the Paris hotel in Las Vegas. Most of the spectators were in town for the DEF CON hacker conference starting Friday at the same site.
The Electronic Frontier Foundation wrote "We think that this initiative by DARPA is very cool, very innovative, and could have been a little dangerous." Sharing their blog post about automated security research, the EFF's staff technologist Peter Eckersley writes: EFF is asking, does research like that need a safety protocol?

not first

[phantomfive]

never before played by programs running on supercomputers.

Except of course last year for the same event.......

a safety protocol?

[fustakrakich]

What good does that do against people who follow no rules?

I'm Sorry Dave

[Jesse Enjaian]

A) It's not that intelligent and certainly doesn't reason. There was a framework to the competition, so it wasn't a wild goose chase within the x86 ISA. For example, https://github.com/CyberGrandChallenge/samples/blob/master/examples/CADET_00001/src/service.c [github.com]. This code has a 'bug" that's only found by brute forcing ASCII characters on input to that function.

B) EFF doesn't need to worry about this thing becoming sentient and starting Skynet [youtube.com]. It's certainly a predefined set of common system flaws which, at best, use machine learning for their implementation.

https://www.cybergrandchallenge.com/tech [cybergrandchallenge.com]