Preventing model exfiltration with upload limits

Unlike most files you might want to secure, model weights are extremely big. This might make them much easier to secure.

Preventing model exfiltration with upload limits Unlike most files you might want to secure, model weights are extremely big. This might make them much easier to secure. Ryan Greenblatt May 08, 2024 3 Share At some point in the future, AI developers will need to ensure that when they train sufficiently capable models, the weights of these models do not leave the developer’s control. Ensuring that weights are not exfiltrated seems crucial for preventing threat models related to both misalignment and misuse. The challenge of defending model weights has previously been discussed in a RAND report

related reading

• Defending Against Model Weight Exfiltration Through Inference Verificationtechnicallyprivate.substack.com
• The case for ensuring that powerful AIs are controlled — LessWronglesswrong.com
• Self-Fulfilling Misalignment Data Might Be Poisoning Our AI Modelsturntrout.com
• Self-exfiltration is a key dangerous capabilityaligned.substack.com
• Efficient LLM inferencefinbarrtimbers.substack.com
• Scaling is subtler than it seemsberen.io
• The case for ensuring that powerful AIs are controlledblog.redwoodresearch.org
• A basic systems architecture for AI agents that do autonomous research — LessWronglesswrong.com
• Compression and Intelligencegreene.sh
• Quantization from the ground upngrok.com
• [2312.06942] AI Control: Improving Safety Despite Intentional Subversionarxiv.org
• AI Control: Improving Safety Despite Intentional Subversion — LessWronglesswrong.com