Incident Overview
On February 6, 2023, a sequence of devastating earthquakes struck southeastern Türkiye, centered near Kahramanmaraş. The first quake registered a magnitude of 7.8, followed by a second of magnitude 7.5, resulting in over 55,000 fatalities and widespread destruction. During this event, Google’s Android Earthquake Alerts system, designed to provide early warnings via smartphones, severely underestimated the magnitude of the initial quake.
The failure of the system led to critically delayed or absent warnings. Only 469 users received the highest-level ‘Take Action’ alert, which urges immediate evacuation. Approximately 500,000 users received lower-tier notifications, while an estimated 10 million people in high-risk zones could have benefited from urgent warnings had the system functioned correctly. These warnings could have provided up to 35 seconds of critical advance notice, potentially saving lives by allowing individuals to seek shelter or take protective measures.
Google’s Android Earthquake Alerts system relies on crowdsourced data from Android devices to detect seismic activity in real time. However, the system’s performance during the Kahramanmaraş earthquakes revealed critical flaws in its magnitude estimation algorithm, particularly in rapidly evolving seismic events.
The incident was first reported by BBC Newsnight in 2025, following an investigation into the system’s failures. Google officially acknowledged the shortcomings of its earthquake alert system in a statement, confirming that the underestimation of the quake’s magnitude directly contributed to the inadequate dissemination of warnings. This case remains the largest known failure of an AI-driven early-warning system to date.
The delayed or absent alerts contributed to the high death toll, as many individuals were caught off guard by the quake’s intensity. The incident underscored the critical role of accurate and timely earthquake warnings in mitigating disaster impacts, particularly in regions with high seismic activity.
Google’s admission of the failure came two years after the event, following sustained media scrutiny and research into the system’s performance. The company published updates to its algorithm in response, though the incident highlighted broader challenges in deploying AI-driven early-warning systems in real-world disaster scenarios.
Possible Root Cause
The primary root cause of the Android Earthquake Alerts system’s failure during the Kahramanmaraş earthquakes was its inability to accurately estimate the magnitude of the initial quake. The system’s algorithm, which processes data from Android devices to detect seismic activity, struggled to interpret the rapid and intense seismic signals generated by the magnitude 7.8 event. This led to an initial magnitude estimate of 4.5–4.9, far below the actual severity of the quake.
This discrepancy in alert prioritization further reduced the system’s effectiveness in a life-threatening scenario.
This absence of redundancy left the system vulnerable to errors, particularly in regions where smartphone penetration or network connectivity might be inconsistent.
The incident also revealed broader challenges in deploying AI-driven early-warning systems in high-stakes environments.
Takeaways
The failure of Google’s Android Earthquake Alerts system during the Kahramanmaraş earthquakes serves as a critical case study in the risks associated with AI-driven early-warning systems. For organizations developing similar systems, the case highlights the need for rigorous testing against a diverse set of seismic events, including rare but high-impact occurrences.
One key takeaway is the importance of redundancy and cross-validation in AI-driven warning systems. The Android Earthquake Alerts system’s reliance solely on crowdsourced smartphone data left it susceptible to errors, particularly in rapidly evolving events.
The system’s static approach to magnitude estimation proved inadequate for the Kahramanmaraş earthquakes, where seismic activity unfolded in ways the algorithm was not designed to handle. Future iterations of such systems must incorporate machine learning models capable of real-time adjustment, learning from incoming data to refine predictions continuously.
The Android Earthquake Alerts system’s tiered warning structure, while useful in less severe events, failed to provide actionable guidance during the Kahramanmaraş quakes. Only the highest-level ‘Take Action’ alert prompted immediate evacuation, yet this alert was not triggered due to the magnitude underestimation. Revisiting alert thresholds and ensuring that lower-tier warnings still convey urgency could mitigate similar failures in the future.
The case also raises questions about the transparency and accountability of AI systems in public safety applications. Google’s acknowledgment of the failure came only after extensive media scrutiny and research, highlighting a potential gap in proactive disclosure. Organizations deploying AI-driven early-warning systems should establish clear protocols for incident reporting and public communication, ensuring that failures are addressed promptly and transparently.
For policymakers and regulators, the incident serves as a reminder of the need for robust oversight of AI systems in safety-critical domains. The Android Earthquake Alerts system was deployed without sufficient scrutiny of its algorithmic limitations, despite its potential to impact millions.
Finally, the Kahramanmaraş earthquake incident highlights the ethical responsibilities of organizations developing AI systems with public safety implications. The potential to save lives through early warnings comes with the obligation to design, test, and deploy these systems with the utmost care. Neglecting to account for edge cases or rare but high-impact scenarios can have devastating consequences, as demonstrated by the failure to warn millions during the quakes.
Sources
- bbc.comhttps://www.bbc.com/news/articles/cx2gq3yqx0po
- incidentdatabase.aihttps://incidentdatabase.ai/cite/1546/
Incident summary from the AI Incident Database, used under CC BY-SA 4.0.