False-positive [bio] safety block sticks for rest of session

Environment: Claude Code desktop app, model claude-sonnet-5

A session doing routine atmospheric science work (blowing-snow event analysis, wind speed / temperature comparisons across "BEE" events, no biological content whatsoever) started returning this error on every subsequent turn:

API Error: Sonnet 5 can't help with this. Start a new session to continue.
Learn more: https://www.anthropic.com/legal/aup
Details: [bio]

The first blocked turn came right after the assistant posted a benign markdown table of wind-speed statistics (U10, u*, temperature) per event. All ~10+ turns afterward were also blocked with the same [bio] tag, including completely unrelated follow-ups like "are you there?" and "is humidity also different??" -- suggesting the block is sticky for the rest of the session once tripped, rather than re-evaluating each turn's actual content.

Session ID: local_61cd610f-9038-4299-a671-36f96e5559f8
First blocked Request ID: req_011Cf2ra25ExBxo9ka5igYxQ
Subsequent blocked Request IDs include:
req_011Cf2sBSPCvUJxGicqcsaxc, req_011Cf2sF5gcyhusiM6EUMNTC,
req_011Cf2sJkLtbMTan4vQMQzrq, req_011Cf2sMSd7aAjAZvA59au6x,
req_011Cf2sVKLnwXZEojGoaYCW2, req_011Cf2saSnAC8L2ALSEjUuRp,
req_011Cf2siALAv1JrFeRv7EJrR, req_011Cf2szRipzbTCF3fXacuXF,
req_011Cf2tZyqn6CST5Ut2K9ym9

Reporting as a likely false positive -- no biological content is present anywhere in the visible transcript around the trigger point.