AI Majority Stop Reason
Updated: June 15, 2026
Description
A majority of AI model responses during a specific timeframe were terminated before natural completion.
This was primarily due to either reaching the maximum token limit or being halted by system guardrails. Such behavior may limit the usefulness of responses or indicate overly constrained generation settings.
Remediation
Refine prompts to be more concise in order to reduce token usage. Review guardrail logs to identify and address any blocked content. If necessary, adjust generation token limits or guardrail policies to allow for more complete responses.
Security Frameworks
Technically allow for the automatic recording of events ('logs') over the lifetime of the system to ensure traceability of functioning appropriate to the intended purpose; for Annex III(1)(a) systems, logs must include the period of use, reference database, input data leading to a match, and identification of natural persons involved in result verification.
Achieve appropriate levels of accuracy, robustness and cybersecurity, and perform consistently in those respects throughout the lifecycle. Declare accuracy levels and relevant metrics in instructions for use. Implement technical/organisational measures against errors, faults, inconsistencies, feedback loops (in continuously learning systems), and adversarial attacks such as data/model poisoning, adversarial examples, model evasion, confidentiality attacks and model flaws.
Take appropriate technical and organisational measures to use the system in accordance with the instructions for use; assign human oversight to competent, trained, supported natural persons; ensure input data is relevant and sufficiently representative for the intended purpose (to the extent the deployer controls data); monitor operation; suspend use and inform provider/distributor/authorities where risk under Art 79(1) is identified or after a serious incident; keep automatically generated logs for ≥6 months; inform workers and representatives prior to workplace deployment; comply with GDPR DPIA obligations; for law-enforcement use, register in EU database; inform persons subject to decisions; cooperate with authorities.
Establish and document a post-market monitoring system proportionate to the nature of AI technologies and risks; actively and systematically collect, document and analyse data on performance throughout the lifetime of the high-risk system; evaluate continuous compliance with Section 2 requirements. Implement based on a post-market monitoring plan (template to be provided by the Commission).
The functionality and behavior of the AI system and its components - as identified in the MAP function - are monitored when in production.
The organization shall define and document the necessary elements for the ongoing operation of the AI system. At the minimum, this should include system and performance monitoring, repairs, updates and support.
The organization shall determine at which phases of the AI system life cycle, record keeping of event logs should be enabled, but at the minimum when the AI system is in use.
A single fault (hallucination, malicious input, corrupted tool, or poisoned memory) propagates across autonomous agents, compounding into system-wide harm.