WH, Anthropic said to draft AI security standards

 The White House and Anthropic PBC are discussing new standards for evaluating security flaws in advanced artificial intelligence models, Politico reported on Thursday, citing administration officials.

Officials are seeking a common method for assessing vulnerabilities known as "jailbreaks," which allow users to bypass an AI model's safeguards. According to the report, the framework could help determine the severity of future incidents and when government intervention may be necessary.

The talks come as Anthropic seeks to lift restrictions on its Fable 5 and Mythos 5 models. The discussions follow a dispute over a security flaw in the company's latest AI systems that led Washington to impose export controls and restrict access for foreign users.

https://breakingthenews.net/Article/WH-Anthropic-said-to-draft-AI-security-standards/66536272