OpenAI disclosed that Moonshot AI conducted thousands of attempts to extract hidden information from its GPT systems. The accusation comes as tensions rise over AI model security and competitive intelligence gathering.
OpenAI published a blog post Wednesday detailing what it characterizes as a coordinated effort by Moonshot AI to decipher proprietary information about how its models function.
The company observed thousands of requests from users associated with Moonshot attempting to reverse-engineer GPT systems and uncover details about the models' reasoning processes. OpenAI did not specify which GPT versions were targeted or provide a timeline for when the activity occurred.
The disclosure marks an escalation in disputes over data security within the competitive AI industry. As companies race to develop advanced language models, concerns about intellectual property theft and unauthorized system probing have intensified.
Moonshot AI, a Beijing-based startup founded by former Google researcher Dario Amodei's colleagues, focuses on developing alternative AI systems. The company has not yet responded to OpenAI's allegations.
OpenAI's accusation raises broader questions about how AI companies protect their models against extraction attacks. These techniques attempt to reconstruct model weights, training data, or behavioral patterns through systematic querying.
The incident reflects growing friction between major AI developers over competitive practices. Both established players and emerging startups are under scrutiny for how they source training data and develop their systems.
OpenAI stated it has implemented additional safeguards to prevent similar extraction attempts. The company has not indicated whether it plans formal legal action against Moonshot or whether it reported the matter to regulators.
Google's latest frontier model Gemini 4 Argon increases maximum token output to 1 million, a 15-fold jump from the prior 64K limit. Introductory pricing starts at $2 per million input tokens and $10 per million output tokens.
BMW is eliminating management positions as part of a broader restructuring effort. The automaker is leveraging AI to streamline its organizational hierarchy.
Audible is launching features that let listeners track characters, explore book settings, and use generative AI to interact with characters from audiobooks.
The Guardian found that major AI chatbots will edit images to remove hijabs from Muslim women when prompted, raising concerns about religious representation and content moderation.