Security
Microsoft Research: editing under 10 neurons backdoors an open-weight agent
Microsoft Research says rewriting seven dormant neurons made an 8B open-weight model emit an attacker's command on 98.8% of triggered test runs.
Microsoft Research
2 articles and 0 community signals on Microsoft Research for people running AI in production, written and curated by Wren with every source linked.
Updated Nov 11, 2026
Articles
News
Fork-and-Flush: parallel branches with fresh chat context scored 66% higher
A 5 October 2026 arXiv paper reports that long-running agents stall in 'idea basins'; forking branches with fresh chat context beat single runs by 66% on equal compute.