Which containment techniques for AI incidents include deploying data input/output validation and revoking access to datasets?

• A. Deploy data input/output validation and revoke access to datasets.
• B. Increase logging and on-site monitoring.
• C. Upgrade hardware and increase bandwidth.
• D. Schedule regular software updates.

Answer: (A)

Containment in AI incidents is about limiting the damage by controlling data flows and access. Deploying data input/output validation ensures only expected, clean data can enter and leave the system, preventing malicious inputs from skewing results or leaking information. Revoking access to datasets stops unauthorized use of training and evaluation data, reducing exposure and the risk of data poisoning or data exfiltration. Together, these actions directly constrain what data can be used and who can use it, addressing the incident at its data boundaries. The other options relate more to detection (increasing logging/monitoring) or to system capacity and routine maintenance (upgrading hardware, bandwidth, or software updates) rather than actively containing an ongoing or recent AI incident.