White House set to finalize voluntary AI jailbreak standards
Anthropic, OpenAI, Google, Microsoft and Amazon are expected to adopt a shared severity scale and 30-day pre-release review by early August.
MVMara Vidović
Published 8 Jul 2026, 15:14Updated 26 Jul 2026, 02:481 min read
- ▸Five labs — Anthropic, OpenAI, Google, Microsoft and Amazon — are finalizing voluntary AI safety standards
- ▸The framework includes a 30-day pre-release review period
- ▸A shared jailbreak severity scoring system is being adopted for the first time across labs
- ▸The push follows a June 2 Executive Order on AI and cybersecurity
- ▸Formal announcement is expected as early as the first week of August 2026
The first shared scoring system for jailbreaks
A common severity scale across five major labs would be a first for the industry — until now, each lab has assessed and disclosed jailbreak risk on its own terms, making cross-lab comparison effectively impossible.
Source: Tech Times
