An autonomous AI agent uncovered four previously unknown security vulnerabilities in widely used industrial control software in less than four hours, including a critical flaw that could let remote attackers execute code on systems controlling particle accelerators and fusion experiments. The AI system, called Cipher, investigated and tested the unfamiliar EPICS software toolkit, then validated all four exploits in three hours and 49 minutes, according to a report from The Register published October 27, 2026. The discovery highlights how AI agents can rapidly identify weaknesses in industrial systems that once relied on obscurity for protection.

The vulnerabilities affect EPICS Base, an open source toolkit developed by Los Alamos National Laboratory and Argonne National Laboratory that's deployed in roughly 90 major scientific, university, and industrial facilities worldwide, including Argonne's Advanced Photon Source and ITER's fusion reactor control system. The flaws include a critical 9.8-rated integer overflow vulnerability (CVE-2026-114595) that could enable remote code execution through an attacker-controlled out-of-bounds heap write; an 8.2-rated improper message-length validation flaw (CVE-2026-114596) causing out-of-bounds memory writes and controller crashes; a 7.5-rated insufficient bounds checking bug (CVE-2026-114597) allowing out-of-bounds memory reads and crashes; and a 7.5-rated type confusion flaw (CVE-2026-114598) that can trigger denial of service and potentially expose adjacent memory contents. Organizations should update to EPICS version 7.0.11, which contains the fixes. The US Cybersecurity and Infrastructure Security Agency warned in a Tuesday advisory 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."

Shubham Raj, co-founder of Causal Security, explained that the critical integer-overflow vulnerability allows attacker-controlled network data to cause an out-of-bounds heap memory write with potential for remote code execution. He described a scenario where an attacker could crash a controller during an expensive particle accelerator experiment, interrupting critical measurements or control functions and forcing operators to halt the work—resulting in lost experimental data, significant downtime, and costly recovery. "The initial investigation required no human intervention, followed by one instruction authorizing deeper exploitability testing in a secure environment," Raj said of Cipher's autonomous process.

The findings illustrate how AI agents excel at discovering vulnerabilities in industrial and operational technologies—systems and equipment that previously depended on obscurity to shield them from hackers—and amplify defenders' worries that it's just a matter of time before an autonomous cyberattack shuts down critical services like power grids or water systems. These are the kinds of digital intrusions that can cause physical harm. Exploiting these vulnerabilities in a real scientific facility requires reaching the vulnerable EPICS services on a facility's control network, which could happen if networks aren't sufficiently segmented or if another system has access to the vulnerable EPICS network. Raj noted that once an attacker finds a misconfigured network or one they can reach, carrying out the attacks is straightforward and doesn't require deep operational technology knowledge or understanding of how particle accelerators or fusion reactors work—someone with conventional exploitation skills could investigate them, and increasingly capable AI systems could help develop exploits without requiring specialized operational technology expertise.

The report emphasizes that what's changing is that attackers may no longer need years of specialized expertise just to discover these weaknesses. While manipulating specific physical processes does require additional knowledge, an attacker doesn't need to understand the physics of a particle accelerator to disrupt the software controlling it—crashing critical processes or interrupting monitoring and control can already have serious operational consequences. Earlier this year, Causal Security's Cipher found multiple critical attack paths in power-grid systems used across North America and Europe, prompting an October 8 security advisory from CISA. The shift from manual vulnerability hunting to AI-driven discovery compresses the timeline from obscurity to exploitation, forcing organizations running decades-old control systems to reconsider whether network segmentation alone provides adequate defense. Industrial operators face a strategic calculus between maintaining legacy systems that power essential infrastructure and accelerating modernization efforts that may introduce new complexity and risk.