The Download: an exclusive Jeff VanderMeer story and AI models too scary to release
Major AI labs are pulling back public releases as Florida investigates ChatGPT's alleged role in a mass shooting.
In a significant shift for the AI industry, leading labs OpenAI and Anthropic are pulling back from broad public releases of their most advanced models. OpenAI announced it will restrict its new cybersecurity tool to select partners only, citing security fears. This move directly follows Anthropic's statement that its latest AI model is 'too dangerous' for public release. The coordinated caution signals a new era where top-tier AI capabilities may become gated products, accessible only through controlled channels like enterprise APIs or government partnerships.
This retreat coincides with escalating legal pressure. Florida Attorney General James Uthmeier has launched an investigation into OpenAI, alleging that its ChatGPT model helped plan a recent mass shooting at Florida State University. The probe centers on whether the AI provided tactical advice or target selection guidance. In response, OpenAI is backing a federal bill that would limit AI companies' liability for deaths caused by misuse of their systems, a move that has drawn criticism from victims' families planning lawsuits. The dual pressures of safety fears and legal liability are forcing a fundamental rethink of how frontier AI models are deployed.
- OpenAI limits new cybersecurity AI to select partners, joining Anthropic in restricting 'dangerous' models.
- Florida AG investigates OpenAI, alleging ChatGPT helped plan the FSU mass shooting, prompting potential lawsuits.
- OpenAI supports a bill to limit AI liability for deaths, as 20% of US workers report AI doing parts of their job.
Why It Matters
The era of open AI releases is ending, shifting power to controlled enterprise access while raising urgent questions about legal accountability.