Ai-driven Worm Emerges as a Potentially Uncontainable Cyber Threat

WHO:

Gizmodo - Researchers from the University of Toronto, the University of Cambridge, and other institutions.

WHAT:

Gizmodo - The researchers' AI worm is described as a "fundamentally new threat" that can autonomously spread and exploit vulnerabilities without human intervention.

Gizmodo - The AI worm was able to infect half of the devices in the experimental network in about five days, but this timeframe is expected to decrease as devices become more efficient at inference and AI models improve.

Fortune - The AI worm can read fresh, publicly available vulnerability advisories online in real time, allowing it to autonomously exploit new flaws without being limited by any LLM knowledge cutoff.

Gizmodo - The researchers' AI worm generates tailored attack strategies to each target it encounters, marking a significant advancement in autonomous malware capabilities.

Fortune - The AI worm can read fresh, publicly available vulnerability advisories online in real time, allowing it to autonomously exploit new flaws without being limited by any LLM knowledge cutoff.

Fortune - The researchers noted that the AI worm took about five days to infect half of the devices in the experimental network, but this timeframe is expected to decrease as devices become more efficient at inference and AI models improve.

Gizmodo - The researchers' AI worm was able to break into nearly three-quarters of the machines on a simulated 33-machine corporate network within one week, establishing a permanent presence on nearly two-thirds of them, with zero human involvement.

Fortune - The AI worm can read fresh, publicly available vulnerability advisories online in real time, allowing it to autonomously exploit new flaws without being limited by any LLM knowledge cutoff.

Fortune - The researchers ran the AI worm 15 times on a simulated 33-machine corporate network, and on average, it broke into nearly three-quarters of the machines within one week with zero human involvement, establishing a permanent presence on nearly two-thirds of them.

Fortune - The AI worm demonstrated the ability to read fresh, publicly available vulnerability advisories online in real time, allowing it to exploit new flaws autonomously.

Gizmodo - The AI worm is capable of dynamically detecting security flaws unique to each device it infects, using a variety of tactics to propagate through a network.

Gizmodo - The worm takes about five days to infect half of the devices in a controlled network, but this timeframe is expected to decrease as devices become more efficient at inference and AI models improve.

Gizmodo - The researchers noted that the worm parasitically feeds off devices' computing power, which is increasingly being built to support computationally expensive large language models (LLMs).

Gizmodo - Researchers demonstrated the creation of an AI-powered worm that can autonomously spread and exploit vulnerabilities in devices.

Gizmodo - The experimental worm took about five days to infect half of the devices in a controlled network.

Techmeme - The worm tailors its attack strategies to each target it encounters.

WHERE:

Gizmodo - The research was conducted in a controlled, isolated network composed of Linux, Windows, and IoT devices.

WHEN:

Gizmodo - A preprint paper outlining the research was published on a Tuesday.

WHY:

Fortune - Gary McGraw, CEO of the AI security nonprofit Berryville Institute of Machine Learning, stated, “This shows what happens when a generic model that’s open weights can be targeted, and it just sort of grinds relentlessly, looking for bugs.”

Fortune - Gary McGraw, CEO of the AI security nonprofit Berryville Institute of Machine Learning, stated, “This is bigger than Mythos in my view,” emphasizing the implications of AI-driven worms that can exploit multiple vulnerabilities.

Fortune - Gary McGraw, CEO of the AI security nonprofit Berryville Institute of Machine Learning, stated, “This is bigger than Mythos in my view,” highlighting the implications of AI-driven worms that can exploit multiple vulnerabilities.

Gizmodo - The researchers aim to alert the global cybersecurity community to the new threat posed by AI-driven malware, emphasizing the need for awareness and preparedness against such threats.

Gizmodo - The researchers aim to alert the global cybersecurity community to the new threat posed by AI-driven malware.

Gizmodo - The worm exploits security vulnerabilities unique to each device, making it a more complex threat than traditional malware.

HOW:

Gizmodo - The researchers consulted with government and scientific bodies to responsibly share their findings without enabling misuse, omitting key methodological details from the published paper to prevent potential exploitation.

Fortune - The AI worm uses reasoning to detect and exploit different vulnerabilities as it spreads, unlike traditional worms that exploit a single vulnerability.

Gizmodo - The worm is powered by an unnamed open-source large language model (LLM).

Gizmodo - It spreads via shared digital connections, such as Wi-Fi networks, and can dynamically detect security flaws in devices.

Gizmodo - The researchers consulted with government and scientific bodies to responsibly share their findings without enabling misuse.

(UPDATED June 4, 2026 23:21 UTC)

Researchers have developed an AI worm that represents a "fundamentally new threat," capable of autonomously spreading and exploiting vulnerabilities without any human intervention. In an experimental network, this AI worm managed to infect half of the devices in just five days, and experts anticipate that this timeframe will shrink as technology advances. Gary McGraw, CEO of the Berryville Institute of Machine Learning, emphasized the importance of responsible sharing of findings, noting that the research team worked with government and scientific bodies to prevent misuse by omitting critical methodological details from their published paper.

(UPDATED June 4, 2026 19:41 UTC)

Researchers have developed an AI worm that generates customized attack strategies for each target, representing a major leap in autonomous malware capabilities. This worm can read real-time vulnerability advisories online, allowing it to exploit new flaws without being hindered by traditional knowledge limitations, and it took about five days to infect half of the devices in an experimental network, a timeframe that is expected to shorten as technology advances. Gary McGraw, CEO of the Berryville Institute of Machine Learning, emphasized the significance of this development, stating, “This is bigger than Mythos in my view,” while the researchers have taken steps to responsibly share their findings to prevent misuse.

(UPDATED June 4, 2026 17:52 UTC)

Researchers have developed an AI worm that successfully infiltrated nearly seventy-five percent of a simulated corporate network of thirty-three machines within just one week, establishing a permanent presence on about two-thirds of them without any human intervention. This advanced AI worm can autonomously read and exploit new vulnerabilities in real time by accessing publicly available advisories online, bypassing the limitations of traditional models. Gary McGraw, CEO of the Berryville Institute of Machine Learning, emphasized that this situation illustrates the risks posed by generic models with open weights, as they relentlessly search for bugs.

(UPDATED June 4, 2026 08:20 UTC)

Researchers have developed an AI worm that, when tested on a simulated 33-machine corporate network, successfully infiltrated nearly three-quarters of the machines within a week, establishing a permanent presence on almost two-thirds of them without any human intervention. This advanced worm can autonomously read and exploit new vulnerabilities in real time by accessing publicly available advisories online, showcasing a significant leap from traditional worms that typically target a single flaw. Gary McGraw, CEO of the Berryville Institute of Machine Learning, emphasized the seriousness of this development, stating, “This is bigger than Mythos in my view,” as it highlights the potential dangers of AI-driven cyber threats.

(UPDATED June 4, 2026 04:21 UTC)

A new AI worm has been developed that can dynamically detect security flaws unique to each device it infects, and it takes about five days to compromise half of the devices in a controlled network, although this timeframe is expected to shorten as technology advances. Researchers are raising awareness within the global cybersecurity community about this emerging threat, highlighting the importance of preparedness against AI-driven malware. To responsibly share their findings, the researchers consulted with government and scientific bodies and intentionally left out key methodological details from their published paper to prevent misuse.

(UPDATED June 4, 2026 01:52 UTC)

Researchers from the University of Toronto, the University of Cambridge, and other institutions have developed an AI-powered worm capable of autonomously spreading and exploiting vulnerabilities in devices. In a controlled network of Linux, Windows, and IoT devices, this experimental worm managed to infect half of the devices within just five days, tailoring its attack strategies to each target it encountered. The researchers aim to alert the global cybersecurity community about this complex new threat, which is powered by an unnamed open-source large language model and spreads through shared digital connections like Wi-Fi networks while dynamically detecting security flaws.