Is it too late to reduce catastrophic AI risk? (news.harvard.edu)

A recent research paper from Harvard Kennedy School's Belfer Center highlights urgent concerns about the potential for catastrophic events arising from artificial intelligence (AI). The study, co-authored by computer scientist Stephen Casper, underscores that while it may not be too late to prevent a large-scale disaster, AI companies and policymakers must act decisively. It highlights threats posed by both closed-weight AI models, which are easier to safeguard, and open-weight models, which lack sufficient protection and present higher risks of misuse. The paper notably mentions the likelihood of major AI-enabled terrorism events, identifying indicators and potential scenarios that could lead to significant harm if left unaddressed. The significance of this research lies in its timely warning, as prominent AI leaders, including the CEOs of Anthropic and OpenAI, advocate for a pause in AI development due to safety concerns. With the proliferation of AI capabilities potentially leading to large-scale incidents, the authors argue that the decisions made by AI firms and governments today will greatly influence future outcomes. The paper emphasizes that while risks are growing—particularly in the realms of bioweapons, cyberattacks, and terrorism—the implementation of robust safeguards and regulatory measures can help mitigate these threats, fostering a more informed and responsible approach toward AI deployment.