Uncle Sam Needs Your AI Help, and DARPA Will Hand Out $20M to Make it Happen

pcmag.com:

LAS VEGAS—The Pentagon’s research-and-development shop showed up at the Black Hat security conference here with a request and a reward: Show us how you can use AI to secure critical software and infrastructure, and you could win a share of almost $20 million in prizes.

The Defense Advanced Research Projects Agency (DARPA) unveiled its AI Cyber Challenge in an unscheduled appearance at the end of the keynote that opened Black Hat this morning.

Perri Adams, manager of this new “AIxCC” program, came onstage to say that DARPA had its eyes not on AI’s downside risks but on its potential upside for security.

"The gains in AI, when used responsibly, have remarkable potential to secure our code," but, she added, such defensive applications of AI as automated scanning and fixing of software vulnerabilities won’t happen without a push of some kind: “We need a forcing function." 

DARPA aims to provide that with the two-year AIxCC program, which will feature a funded track (in which competitors have until September to submit proposals, after which up to seven will receive funding from the agency) and an open track. 

DARPA will then select up to 20 qualifying teams from both tracks for a semifinal competition at the DEF CON security conference here in August 2024, at which DARPA will award the top five competitors $2 million each. A final round at DEF CON 2025 will see the agency pick the top three winners and award additional incentives: $4 million for first place, $3 million for second place, and $1.5 million for third. 

“By automatically defending critical software at scale, we can have the greatest impact for cybersecurity across the country, and the world,” Adams said in a DARPA announcement that also touts support for this