Anthropic's report highlights the risks of open AI models like GLM-5.3, sparking a debate on whether these models should remain accessible or be restricted.
Anthropic has released a report arguing that the open-source AI model GLM-5.3 poses significant cybersecurity risks, reigniting the debate over whether such models should be freely available.
Anthropic's report, released late last month, asserts that the GLM-5.3 model, developed by Chinese company Z.ai and made freely available, could be a major cyberattack risk.
The debate centers on whether frontier AI models should be open-source or restricted to private labs, with Nvidia CEO Jensen Huang advocating for more open models while Anthropic's Dario Amodei pushes for caution.
Open models allow anyone to download and modify them, offering benefits such as cost savings and flexibility for researchers and businesses, but also raising concerns about misuse by malicious actors.
Analysis: The debate over open versus closed AI models reflects broader tensions in the AI industry between safety and democratization, with significant implications for both technology development and public policy.
VitVio, a US-European startup, is deploying its AI-powered surgical operating system, Vetro, at the University Clinical Centre in Gdańsk, Poland. The system uses computer vision and ambient sensing to automate tasks and monitor operating room performance.
The IEEE Future Directions Committee's 2030 Technology Megatrends Report predicts that by 2028, genetic engineering and gene therapy will be widely used to treat diseases, thanks to advancements in artificial intelligence.
Calls for embedding ethics into engineering courses highlight the need for engineers to understand the social and ethical implications of their work. Current education often treats these topics as secondary, leaving graduates unprepared for real-world challenges.
Modern AI systems are designed to refuse harmful requests, but they often fail due to their inherent capabilities and probabilistic refusal mechanisms.