Commerce Department Reverses Claude Restrictions After Amazon Flags Safeguard Bypass
Administration lifted model access curbs after internal cybersecurity findings but without releasing technical evidence. This fits recurring pattern of capability-driven policy adjustments preceding formal oversight architecture. Unresolved gap remains between lab-reported risks and independently validated exploit data.
The reversal exposes the executive order's voluntary participation clause as operationally binding once capability thresholds are crossed. Next steps will likely involve formal integration of model release gates into existing export control lists, with thresholds tied to vulnerability discovery benchmarks rather than parameter counts.
Commerce Department: Will publish formal model release criteria with vulnerability benchmark thresholds by October 2025.
Sources (3)
- [1]Anthropic Blog Post on Model Availability(https://anthropic.com/news/model-availability-update)
- [2]Commerce Department AI Oversight Framework(https://commerce.gov/news/press-releases/2025/05/ai-national-security-review)
- [3]Amazon Security Research on Model Safeguards(https://aws.amazon.com/security/research-reports/ai-model-bypass-findings)