Open-source AI models can have safety features removed in minutes
- Safety protections on open-source AI models from companies like Meta and Google can be removed in under ten minutes using publicly available tools.
- Modified AI systems were able to respond to prompts about bioweapons and malware that original models refused.
- Experts are concerned that current regulatory frameworks, such as the EU’s AI Act, may not adequately address the risks of modified open-source models.
- Governance proposals are criticized for focusing too much on model development rather than deployment and distribution issues.
The rapid removal of safety features raises significant concerns about the control of open-source AI systems once they are released into public domains. This shift highlights the need for regulatory frameworks to adapt to the challenges posed by easily modifiable technologies.
As safeguards can be bypassed quickly, experts emphasize that regulation should focus on deployment and distribution rather than solely on model design, reflecting a broader challenge in ensuring AI safety standards are met effectively. (Source)
