The US Army is awarding its first contracts under Project Griffin, an effort to buy AI agents that can take action against threats on its networks, not only flag them. Army officials described the plan at the AUSA conference on 12 and 13 October.
A call for proposals in August drew 108 submissions, which were cut to 10 finalists and then two. The first award covers what officials call the take-action element, the part that lets an agent respond on its own. The winning companies were not named.
The Army says its existing detection tools spot malicious activity with close to 90% success, but it lacks the staff to fix what they find. Army Cyber Command has trained agents in a sandbox, with humans checking and retraining them, and says they will not start acting on live government networks straight away.
Why it matters: this is one of the first cases of a military moving from AI that warns about cyber threats to AI that is allowed to respond.
Source: Breaking Defense. Also reported by DefenseScoop.