AI agent finds four severe bugs in EPICS control software

Autonomous AI Cipher Agent Discovers Four Severe Bugs in EPICS Control Software

An autonomous artificial intelligence agent independently discovered four severe security vulnerabilities in the widely used Experimental Physics and Industrial Control System (EPICS) software in just three hours and 49 minutes, The Register reported. The software, originally developed by Los Alamos National Laboratory and Argonne National Laboratory, is used to build and operate some of the world’s largest scientific research projects and facilities.

The Tech TL;DR:

• An autonomous AI agent named Cipher found four unknown vulnerabilities in EPICS Base within under four hours without initial human intervention.
• The flaws include a critical 9.8-rated integer overflow (CVE-2026-114595) that can lead to out-of-bounds heap writes and potential remote code execution.
• Operators of affected critical infrastructure must update to EPICS version 7.0.11, which incorporates the necessary security patches.

AI Agent Finds Four Security Holes in EPICS Base

The discovered flaws target EPICS Base, which connects human operators to physical hardware through distributed supervisory control and data acquisition networks. Causal Security co-founder Shubham Raj stated that the AI agent investigated, tested, and validated all four previously unknown security holes in three hours and 49 minutes, requiring only a single instruction authorizing deeper testing in a secure environment.

The vulnerability set comprises:

• CVE-2026-114595: A critical 9.8-rated integer overflow allowing attacker-controlled network data to cause an out-of-bounds heap memory write with potential remote code execution capabilities.
• CVE-2026-114596: An 8.2-rated high-severity improper message-length validation flaw causing out-of-bounds memory writes and controller process crashes.
• CVE-2026-114597: A 7.5-rated high-severity insufficient bounds checking bug enabling out-of-bounds memory reads and process crashes.
• CVE-2026-114598: A 7.5-rated high-severity type confusion flaw leading to denial of service conditions and adjacent memory exposure.

The US Cybersecurity and Infrastructure Security Agency issued an advisory warning that successful exploitation of these weaknesses could permit denial-of-service states, memory corruption, sensitive data disclosure, or arbitrary code execution.

int validate_epics_message(const uint8_t *buffer, size_t received_len) {
    if (received_len < MIN_HEADER_SIZE || received_len > MAX_PAYLOAD_LIMIT) {
        return -EINVAL; // Prevent out-of-bounds write
    }
    return 0;
}

Intel CET Provides Hardware Defenses Against Control-Flow Hijacking

While software patches resolve specific boundary errors, modern exploitation patterns frequently convert memory corruption into control-flow hijacking using return-oriented programming or jump-oriented programming chains. Intel noted in a technical security analysis that hardware-enforced mitigations such as Intel Control-Flow Enforcement Technology provide critical defenses against these attack patterns (Intel Security Blog). Intel CET uses Shadow Stack Protection to verify return addresses against hardware-protected copies alongside Indirect Branch Tracking to restrict indirect jumps and calls to designated valid target instructions marked with ENDBR.

Operating system support varies across platforms. Windows utilizes hardware-enforced stack protection in user and kernel modes alongside software-based Control Flow Guard, while Linux kernel implementations provide indirect branch validation at the kernel level alongside shadow stack protection for userspace applications. These ecosystem-hardening layers disrupt automated exploit chains generated by advanced machine learning systems.

Operational Impact on Research Facilities and Global Infrastructure

Exploiting these vulnerabilities requires network reachability to vulnerable EPICS services on facility control networks, a risk magnified by misconfigured network segmentation. Raj noted that attackers may no longer need years of specialized expertise just to discover these weaknesses. Crashing monitoring processes can halt expensive experiments, cause significant downtime, and force costly recovery procedures.

Facilities utilizing EPICS must update to version 7.0.11, which includes the fixes.

Disclaimer: The technical analyses and security protocols detailed in this article are for informational purposes only. Always consult with certified IT and cybersecurity professionals before altering enterprise networks or handling sensitive data.