OpenAI reports thwarting model reasoning theft, Azure remains vulnerable

VERA Brief

AI-generated. Grounded in the article and its cited sources.

OpenAI disrupted a campaign involving over 15,000 accounts attempting to extract its models' reasoning and linked some activity to Moonshot AI. However, researchers observed the same attack method continuing to affect Microsoft Azure, including newer models, for several weeks.

Key facts

• OpenAI stated it disrupted a campaign to extract its models' reasoning capabilities.
• The company reported that over 15,000 accounts were involved in the campaign.
• OpenAI attributed a portion of the activity to individuals associated with Moonshot AI.
• Researchers observed the same attack method continuing to affect Microsoft Azure for several weeks.
• The attack method included targeting newer iterations of OpenAI's models.