As cited
Copy frozen at (site build).
ai security
Srsly Risky Biz: US Vows to Fight Distillation Attacks
The US government has pledged to counter Chinese distillation attacks that extract capabilities from advanced American AI models by training less capable models on their outputs. OpenAI, Google, and Anthropic previously disclosed being targeted by such attacks, with Anthropic reporting 16 million exchanges across fraudulent accounts and Google citing 100,000 queries to Gemini. The article expresses skepticism about the effectiveness of the government's response.
Why it matters: AI model developers and enterprises relying on frontier AI systems should monitor for signs of model extraction attempts and understand that proprietary model capabilities can be replicated through legitimate API queries, requiring enhanced access controls and usage monitoring.
- Source published
- First seen by Cybersecurity Tracker
Source attribution
Correction
Correction recorded as of .
ai security
Srsly Risky Biz: US Vows to Fight Distillation Attacks
No summary had been written when this copy was frozen.
- First seen by Cybersecurity Tracker
Source attribution
Correction
Correction recorded as of .
ai security
Srsly Risky Biz: US Vows to Fight Distillation Attacks
The United States (US) government has announced it will counter Chinese model extraction, or distillation, attacks aimed at stealing the capabilities of frontier Artificial Intelligence (AI) models. Officials described how the attacks involved tens of thousands of queries to models such as Gemini and Claude, allowing adversaries to cheaply improve their own AI systems.
Why it matters: US AI developers and government agencies face model theft via distillation attacks and should monitor for abnormal query volumes and consider deploying extraction-defense controls.
- Source published
- First seen by Cybersecurity Tracker
Source attribution
Correction
Correction recorded as of .
ai security
Srsly Risky Biz: US Vows to Fight Distillation Attacks
The United States (US) government has announced it will counter Chinese model extraction, or distillation, attacks aimed at stealing the capabilities of frontier Artificial Intelligence (AI) models. Officials described how the attacks involved tens of thousands of queries to models such as Gemini and Claude, allowing adversaries to cheaply improve their own AI systems.
Why it matters: US AI developers and government agencies face model theft via distillation attacks and should monitor for abnormal query volumes and consider deploying extraction-defense controls.
- Source published
- First seen by Cybersecurity Tracker