AI Control: A New Enterprise Resilience Priority
October 1, 2026 - 12:27 pm
Credit: Unleash
Artificial intelligence is deepening its penetration into enterprise infrastructure, introducing a broader range of risks. As AI evolves from generating text in isolated environments to initiating transactions, influencing decisions, and contributing to production code, the consequences of an error can ripple across interconnected systems.
Splunk’s 2026 research, released by Cisco in partnership with Oxford Economics, reveals that unplanned downtime costs Global 2000 companies approximately $600 billion annually, with average downtime costs reaching a staggering $15,000 per minute.
Recent cloud disruptions highlight the impact of a hasty rollout. During a significant Google Cloud outage in June 2025, a new feature was activated globally, according to Google’s subsequent incident report. This rapid deployment led to immediate failures across multiple regions and services, affecting third-party platforms dependent on Google Cloud infrastructure.
Google identified progressive rollouts as a missed safeguard. This incident underscores a critical resilience principle: minimizing initial exposure to change can mitigate its potential impact and provide teams with a chance to intervene before a problem spreads.
This exposure also ignites a broader governance debate. Splunk’s research found that 68% of technology leaders expressed concern about unpredictable AI-agent behavior, and every leader surveyed experienced AI-related downtime.
As Google CEO Sundar Pichai noted in Alphabet’s annual report, "Today, nearly 75% of all new code at Google is AI-generated and approved by engineers, up from 50% last fall." As AI accelerates software development, incidents like Google’s underscore the importance of controls governing the release of changes to production.
These findings suggest that observability alone may not be sufficient. Monitoring can detect unusual behavior, but an intervention mechanism is crucial to stop defined processes when conditions are met.
The question shifts from observing AI systems to controlling them operationally. As automated software becomes more capable and integrated into production, mechanisms for pausing, disabling, or reversing AI-enabled functions should be integrated into resilience planning, alongside monitoring and incident response.
Egil Østhus, CEO of Unleash, frames this challenge within the context of software release management. Unleash is an open-source feature management platform that separates code deployment from the decision to activate or change software behavior, enabling organizations to control applications, services, and AI-enabled capabilities at runtime.