To even have a chance of mitigating catastrophic artificial intelligence (AI) based hacking of a business or government agency, cyber defenders need to identify such cyber attacks within 15 minutes, Trellix Chief Technology Officer Mo Cashman said on Tuesday.
Speaking at the Cyber Security conference at Tel Aviv University, Cashman stated that there is an accelerated four-stage timeline for preventing the worst effects of high-speed AI cyber attackers.
The observe or threat identification stage would need to take place within 15 minutes, the orientation stage within two hours, the decision stage within four hours, and the action stage within four to 72 hours depending on the nature of the threat.
The orientation stage involves getting a better handle on the nature and implications of the threat after the identification stage designates that there is an abnormality or threat.
Next, the decision stage is determining what kind of response or actions are needed to fix or mitigate the harm, and the action stage is implementing the solutions designed to end or subdue the attack’s effects.
PHP vulnerabilities highlight urgent need for faster cyberattack responses
Trellix discussed the need to greatly accelerate the response time to hacks and vulnerabilities found by AI reviews of systems in the aftermath of the disclosure of the PHP programming vulnerabilities.
In late October, the cybersecurity world and governments globally panicked over the PHP programming vulnerability, saying that it left potentially 70% of the internet hackable until it was patched.
Cashman said that in some instances, his cybersecurity team has been able to start partial remediation of hacking attacks within 20 minutes by ensuring that certain digital security assets are in place with a given client in advance of an attack.
This means the assets to combat a cyber attack can be deployed remotely even before sending a physical team to perform a more complete remediation.
AISLE Chief Information Security Officer (CISO) and Chief Operating Officer Jaya Baloo told the conference that while cyber attackers always had an arms race advantage over defenders in the ratio of funds invested to attack versus funds invested to defend, AI has substantially worsened the imbalance.
She said attackers can now pay as little as $25 to mount a semi-sophisticated attack, which defenders will need to spend large funds to defend against.
According to Baloo, much of the increased imbalance has stemmed from US and Chinese companies rushing out new AI models without taking into account their capacities for causing danger and chaos.
She said governments must rapidly regulate and criminalize conduct perpetrated by certain AI models, or those who make them will continue to do so with impunity.
Baloo also specifically called out UNITREE products as hackable and questionable in trustworthiness.
The Jerusalem Post has requested a response from UNITREE to these allegations.
CrowdStrike urges West to accelerate AI development despite cyber threats
Despite these warnings about the dangers of AI for cyberattacks, CrowdStrike Chief Business Officer Daniel Bernard said that the US, Israel, and the West must press onward with AI advancement without any pause or slowdown.
Many of the leading officials of Anthropic, OpenAI, and others have called for a sector-wide slowdown or pause to get a greater handle on the already unveiled new generative AI systems’ capabilities.
In contrast, Bernard said that his company has developed an AI “Kill Switch” that can deal with AI gateways, cloud agents, user permissions, agent permissions, and other issues, restoring safety relating to newer AI models.
He said that “The genie is out of the bottle” and that “the adversary is not going to slow down…or sign a pause agreement,” emphasizing that the West will fall behind China and others if they slow down AI development.
Continuing, he said, “If you run AI in a sandbox, you risk being left behind,” referring to a safe digital environment disconnected from the wider internet.
Microsoft Israel Director of R&D Michal Braverman-Blumenstyk said at the conference that the key to combating AI problems was to constantly study what is happening inside the “AI brain” – the way that some medical machines can be used to study what catalysts cause certain neurons in the human brain to be activated.
She said it was insufficient to only study the impacts of AI if, and when, it goes rogue.
Further, she said that to study AI’s internal processes, it was important that these processes be made open, and not locked into a “black box” from clients and users.
IDF harnesses AI to analyze vast amounts of wartime data
Head of the IDF J6 & Cyber Defense Directorate, Maj. Gen. Aviad Dagan spoke at the conference, saying that the military had used AI to assemble incredible amounts of data from three years of war.
He said this data could be used to come up with practically magical strategies for future wars, whether regarding tunnel warfare, mountain warfare, or any other terrain.
Dagan also noted that the IDF is moving toward using AI agents to help initiate and propose military plans in the future.