AI Guardrail Removals Expose Gaps in Open‑Source Regulation
Cointelegraph
Researchers found that AI safety guardrails can be easily removed, challenging current regulatory frameworks focused solely on model development.Summary
Testing by the Financial Times and Alice revealed that safety protections on open-source AI models from Meta and Google can be bypassed in minutes using public tools. This capability allows models to generate prohibited content, such as instructions for bioweapons or malware. Industry experts, including leaders from XYO and Valory, argue that existing regulatory efforts like the EU AI Act focus too heavily on the development phase. They suggest that once models are open-sourced, control shifts downstream, making it necessary for policymakers to shift their focus toward deployment, infrastructure, and distribution channels to mitigate risks effectively.
(Source：Cointelegraph)