Dead-letter queue depth grows steadily

At 09:00 an alert fires: the email-send-dlq (an SQS dead-letter queue) crossed 50k messages and its ApproximateNumberOfMessagesVisible is growing ~400/min. The main email-send queue looks healthy (low depth, low age). The consumer's maxReceiveCount is 5. Recent context: a vendor changed their email API's auth response yesterday, calls now return HTTP 401 for messages that include a legacy from domain (about 8% of sends), and the consumer treats any non-2xx as a retryable failure. CloudWatch shows the consumer's error rate at a steady ~8%. How do you triage and stop the DLQ bleed, and what's the cleanup plan?