The Race Against AI Why Automation Specialists Need a Head Start with New AI Models

Related Vendors

CISPA Helmholtz Center for Information Security

FAULHABER Drive Systems

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

SIMTOS (Seoul International Machine Tool Show)

New generations of AI are finding vulnerabilities in industrial software at a rapid pace. To maintain system availability, researchers are calling for a paradigm shift: software and automation manufacturers need exclusive early access to new AI models.

Anyone who keeps machines running in industrial manufacturing is familiar with the dilemma: Security patches often require system reboots, which comes at the expense of equipment availability. In the current age of AI, this problem is escalating dramatically. As Christoph Landolt, a researcher at the CISPA Helmholtz Center for Information Security, warns, the number of potential attack vectors multiplies with every release of new AI models.As a prominent example, he cites the launch of “Claude Opus 4.6” in February 2026: In just two days, the model found over 500 serious vulnerabilities in open-source software. According to the researchers, the time between a vulnerability being disclosed and the first cyberattack has shrunk from over two years (in 2018) to around 10 hours today. 

The “Red Queen’s Race” in Automation

For automation engineers, this is like the Red Queen’s race from “Alice in Wonderland”: You have to run just to stay in place. “If attackers discover new vulnerabilities faster than users can update their software, the system begins to oscillate,” explains Landolt. The classic principle of “responsible disclosure” reaches its limits when AI models make both defensive and offensive knowledge accessible to everyone at the same time.
The pressure is enormous, especially in OT. “The defender must not only ensure security but also availability: A vulnerability must be detected, a patch built, tested, and released,” says Landolt. An attacker, on the other hand, doesn’t care if they cause a system shutdown.

Game Theory Provides the Solution: Phased AI Releases

To give industrial software manufacturers a realistic chance to defend themselves again, the CISPA researchers developed a game-theoretic model in their paper “The Oracle’s Gambit.” The conclusion: The AI industry must change its release strategy.
New, powerful “frontier models” must not be made immediately available to the general public (and thus to cybercriminals). Instead, a strict sequencing is required:

1. Exclusive access for software manufacturers and tech companies to identify vulnerabilities and develop patches for control systems and other systems.
2. The AI will not be released to the public until the updates are available. The researchers conclude that only this time buffer can create incentives to effectively patch machines and systems before hackers can exploit the vulnerabilities using automated methods.

Original publication:Landolt, Christoph; Lorenz, Tobias; Kwiatkowska, Marta; Fritz, Mario (2026) “The Oracle’s Gambit: A Game-Theoretic Framework for Responsible AI Release” In: The 17th Conference on Game Theory and AI for Security (GameSec-26), Oct. 26–28, 2026, Ann Arbor, MI, USA
https://doi.org/10.60882/cispa.33086237