OpenAI began a phased release of its newest model Thursday, which the company said surpasses Anthropic’s most advanced models and meaningfully approximates “artificial general intelligence.”
GPT-6 Astra is the startup’s first to cross the “critical” cybersecurity capability threshold, OpenAI said, and will be initially limited to businesses to allow them to address vulnerabilities.
The move echoes Anthropic’s restricted release of Mythos, whose capabilities prompted regulatory scrutiny: The Pentagon affirmed its ban of Mythos Thursday.
OpenAI is striving to regain the initiative in the AI race against a backdrop of growing anxiety from experts, tech leaders, and the public over the risks of frontier AI. Two US lawmakers on Thursday introduced the first, largely symbolic legislation to pause AI development.