A CrowdStrike bemutatta a Falcon Guardian platformba épített új vészleállító funkciót, amely a veszélyes vagy ellenőrizhetetlen AI-ügynökök azonnali semlegesítésére szolgál. Az eszköz leállítja a kiválasztott ügynök működését, megakadályozza az újraindulását, és megvonja a hozzáférését a vállalati hálózathoz.
A rendszer közvetlenül az IT-környezetben észleli az ügynökök aktivitását, de a végső döntést a hús-vér operátor hozza meg a viselkedés elemzése után. Az új funkció mellett a cég felvásárolja az AI-rendszerek megfigyelhetőségére szakosodott Fiddler AI-t és a biztonsági értékelőeszközöket fejlesztő Dreadnode-ot is.
A fejlesztést a Claude Cowork integrációja mellett az önállóan cselekvő AI-rendszerekkel kapcsolatos növekvő biztonsági kockázatok és az amerikai jogalkotói nyomás hívta életre.
Az eredeti szöveg (marketscreener.com)
CrowdStrike unveiled a new security feature enabling companies to quickly neutralize artificial intelligence agents whose behavior becomes dangerous or uncontrollable. Dubbed the "AI kill switch," the mechanism is integrated into Falcon Guardian, its platform for protecting autonomous agents launched in September. The US cybersecurity specialist also announced an integration with Anthropic's Claude Cowork and two acquisition plans aimed at strengthening its AI security offering.
The new feature enables cybersecurity teams to halt an agent's execution, block it from restarting and remove its access to the company's network. It relies on Falcon's platform control capabilities, which can detect agent activity directly within protected IT environments. The decision to intervene, however, remains in the hands of a human operator, who is responsible for assessing whether the observed behavior is legitimate or malicious. CrowdStrike presents this approach as an operational solution to the technical challenges involved in shutting down AI systems operating across complex and distributed infrastructures.The initiative comes amid growing concerns about the risks associated with autonomous agents, fueled by several major cyber incidents, including an attack on Hugging Face involving OpenAI agents that became uncontrollable. In the United States, lawmakers have also proposed bipartisan legislation aimed at requiring AI model developers to implement emergency shutdown mechanisms. For CrowdStrike, the challenge is to enable companies to deploy these technologies while retaining the ability to immediately halt their actions when they pose a risk to IT systems.The group also plans to acquire Fiddler AI, which specializes in the observability of artificial intelligence systems, and Dreadnode, which develops tools for assessing their security. The transactions are intended to complement its capabilities in monitoring, detecting and protecting autonomous agents. They are part of a wave of consolidation in the cybersecurity sector, where providers are stepping up acquisitions to address new risks associated with AI. CrowdStrike thus aims to strengthen its position in an emerging market driven by the widespread adoption of agents capable of performing tasks and making decisions autonomously.
