AI agents could help repel cyberattacks… faster than humans.
At the beginning of the week, the “AI Table Top Exercise 2.0” (AI TTX), the second simulated AI exercise, was organized for the US Army by the independent Strategic Competitive Studies Project.
Representatives from the United States Cyber Command, the Army Cyber Command, and executives from companies such as Amazon Web Services, Google, OpenAI, Microsoft, CrowdStrike, Palo Alto Networks, Darktrace, SentinelOne, and Booz Allen Hamilton participated in the initiative.
The exercise, set in a 2027 Indo-Pacific conflict scenario, prompted military leaders to reflect on new risk management policies and a possible expansion of the autonomy granted to AI agents in cyber operations.
AI advances are forcing military leaders to rethink their defense strategies, and they cannot afford to proceed at typical bureaucratic speeds, comments Axios.
All the details.
THE SIMULATION
The war simulation presented participants with a dramatic scenario: it is 2027, and a crisis in the Indo-Pacific has turned into a cyber war against US military networks.
Much of the discussion focused on how the latest artificial intelligence models are accelerating vulnerability discovery and how these same tools could be used to defend Army systems, Axios further explains.
According to Brandon Pugh, chief cybersecurity advisor to Army Secretary Daniel Driscoll, “the premise was that an adversary was leveraging artificial intelligence to launch a series of attacks that continuously adapted to the Army’s defensive posture, and presumably did so faster than a human defender could.”
THE ACCELERATION DRIVEN BY ARTIFICIAL INTELLIGENCE
According to military officials, the Army used the exercise to better understand how AI could protect its networks, not how it might be employed offensively on the battlefield.
“We are in the early stages of understanding how to defend the AI we are using,” Lieutenant General Christopher Eubank, commander of Army Cyber Command, told reporters. “What we thought would probably be 12 to 18 months away has arrived today, regarding AI and agentic AI.”
“It’s not just about augmenting the human,” Eubank added. “If that were the case, we would be further behind than we think.”
GREATER AUTONOMY FOR AI AGENTS?
The head of Army Cyber Command also explained that new AI tools, which operate largely autonomously with minimal human oversight, can identify cyber vulnerabilities and attack them faster than humans can patch them.
“In this new world of agentic artificial intelligence… telling someone to ‘apply patches faster’ is simply unrealistic,” Eubank told reporters. “So, how can AI be used to get to a point where we are no longer limited by human speed?” he asked. “Where can AI act autonomously in the cyber defense environment?”
AI READY TO INTERVENE IN CYBERATTACKS, NOT JUST HELP DETECT THEM
Artificial intelligence is already helping detect intrusions in Department of Defense networks, explained Brandon Pugh, chief cybersecurity advisor to Army Secretary Daniel Driscoll, but what is now needed is an agentic AI capable not only of alerting human operators to an ongoing attack but also autonomously intervening to stop the breach.
“We are very good at leveraging AI for detection,” Pugh said, reported Breaking Defense. “How can we now continue to advance with agentic capabilities that not only detect but also enable response?”
“Should the degree of human involvement vary depending on the situation we are in?” Pugh continued. “If we are facing a series of cyberattacks during a conflict period, perhaps the risk tolerance is different compared to normal peacetime.”
