An autonomous AI agent identified four security weaknesses within the widely utilized, aging EPICS open source toolkit.
These flaws include one critical vulnerability and three high-severity issues in EPICS. The critical bug, rated 9.8 under CVSS v3.1, is an integer overflow vulnerability (CVE-2026-114595). This flaw permits an attacker to achieve an out-of-bounds heap write, with the potential for remote code execution.
Other identified issues are an 8.2-rated, high-severity improper message-length validation flaw (CVE-2026-114596), which results in an out-of-bounds memory write and controller-process crash. Additionally, there is a 7.5-rated, high-severity insufficient bounds checking bug (CVE-2026-114597) allowing out-of-bounds memory reads and process crashes. Finally, a 7.5-rated, high-severity type confusion flaw (CVE-2026-114598) could cause denial of service or expose adjacent memory contents.
All four vulnerabilities impact EPICS Base, a component essential for operating numerous large scientific research facilities. Organizations are advised to upgrade to version 7.0.11, which contains the necessary patches.
The US Cybersecurity and Infrastructure Security Agency (CISA) issued a Tuesday advisory, stating that "Successful exploitation of these vulnerabilities could allow an attacker to cause a denial-of-service condition, corrupt memory, disclose sensitive information, or potentially execute arbitrary code."
Causal Security co-founder Shubham Raj reported that the AI agent, named Cipher, discovered all four previously unknown security holes. This investigation and validation of the exploits took three hours and 49 minutes. Raj noted, "The initial investigation required no human intervention, followed by one instruction authorizing deeper exploitability testing in a secure environment."
EPICS, which originated from Los Alamos National Laboratory and Argonne National Laboratory, is used to build and remotely manage distributed Supervisory Control and Data Acquisition (SCADA) systems. It is deployed across approximately 90 major scientific, university, and industrial projects globally, including ITER's fusion reactor control system and Argonne National Laboratory's Advanced Photon Source.
These findings underscore the capability of AI agents to locate weaknesses in industrial and operational technologies, which historically relied on obscurity for protection. This heightens concerns among institutional defenders regarding the timeline until an autonomous cyberattack disrupts critical services.
The prerequisite for exploiting these vulnerabilities in a scientific setting is network access to the vulnerable EPICS services. Raj noted that this access could be gained if networks lack proper segmentation or if another system connects to the vulnerable EPICS network.
The discovery highlights a broader trend where sophisticated AI tools are reducing the specialized expertise required for discovering weaknesses in operational technology systems.
