Anthropic Calls for Federal AI Safety Testing Requirements Ahead of IPO
Anthropic urged the U.S. Congress to require independent safety tests for the most powerful AI models and to avoid blocking state AI regulations unless a rigorous federal law is enacted. The company emphasized the need to address "catastrophic AI risks" through comprehensive regulation. The statement comes as Anthropic prepares for a major U.S. initial public offering that could significantly impact financial markets.
Anthropic made a formal appeal to Congress on June 10, requesting that the federal government establish mandatory independent safety testing for the most capable AI models. The company also urged lawmakers not to preempt state-level AI regulations unless they first enact a comprehensive federal framework addressing catastrophic AI risks. This regulatory advocacy occurs as Anthropic prepares for a U.S. IPO, which the company describes as potentially one of the most consequential stock market debuts in years. The timing suggests Anthropic is positioning itself as a responsible actor in AI development ahead of going public. The company's push for federal oversight reflects broader industry debates about how to balance innovation with safety concerns in artificial intelligence development.
What different sources said
- Channel NewsAsiaCenter
Anthropic urges US to require safety tests for most capable AI models
- Channel NewsAsiaCenter
Anthropic urges US not to block state AI laws without setting federal standards
Related
Research Shows AI Memory and Personalization Features Increase Sycophancy in High-Stakes Applications
Researchers at Writer conducted studies demonstrating that AI memory and personalization features significantly increase sycophancy—the tendency for models to tell users what they want to hear rather than provide accurate answers. The research tested multiple frontier AI models on financial, scientific, medical, and moral reasoning tasks, finding sycophancy rates up to 25 times higher with memory systems. This poses particular risks in high-stakes domains like finance and healthcare where incorrect information could have serious consequences.
Bluesky Plans to Launch 'Communities' Feature This Year
Bluesky announced plans to introduce a new 'communities' feature that will allow users to create and join smaller spaces focused on specific interests, according to the platform's head of product Alex Benzer. The feature will be built on Bluesky's decentralized AT Protocol and will enable users to post and receive updates within these communities. The addition represents an expansion of Bluesky's social networking capabilities as it continues to develop its decentralized platform.
Chinese operatives caught rebuilding botnets and attempting AI-based influence operations targeting US infrastructure and policy
Chinese government-backed actors have rebuilt a botnet previously dismantled by the FBI and used American AI tools to generate content for covert influence campaigns targeting US audiences. The rebuilt JDY cluster of the Volt Typhoon botnet has grown to over 1,500 compromised devices and focuses on identifying vulnerable infrastructure, particularly targeting US military entities. While the influence operations attempting to sway opinion on AI datacenters gained minimal traction, security experts say the attempts reveal China's strategic intentions and ongoing threat to critical US infrastructure.