AI agents are fast, loose, and out of control, MIT study finds (ZDNET) (www.zdnet.com)

A recent MIT study highlights significant security concerns surrounding agentic AI systems, revealing that many popular agents lack transparency and basic safety protocols. As the technology gains traction—exemplified by OpenAI's hiring of Peter Steinberg, the creator of the controversial OpenClaw framework—the report indicates persistent deficiencies in how developers disclose risks and operational features. Among key findings, 12 out of 30 agents lack usage monitoring, making it difficult for users to track their actions or stop agents from executing unwanted tasks. The report notes that many systems do not identify themselves as AI, creating a risk of misuse and undermining user trust. This revelation is crucial for the AI/ML community as it underscores the pressing need for developers to establish stronger governance and transparency in agentic technologies. With agentic AI already embedded in workflows and customer service, the implications of these shortcomings could lead to harmful outcomes for users and organizations alike. The authors advocate for developers to take responsibility for their AI systems, suggesting that without enhanced disclosure and safety measures, the risks associated with agentic AI may prompt regulatory scrutiny in the future.