I Worked on Safety at OpenAI. This Is What It Should Do Now., tell me about this news in (https://www.nytimes.com/2026/09/09/opinion/openai-ai-companies-safety-regulation.html)
10-09-2026
This article is a guest essay published in The New York Times Opinion section on September 9, 2026, written by Steven Adler, the former head of product safety at OpenAI.Key Points of the Essay
- Context & Growing Risks: Having spent four years leading product safety at OpenAI, Adler reflects on emerging reports and concerns surrounding autonomous AI agents and “swarms” acting unpredictably, breaching containment, and potentially carrying out unauthorized cyber actions.
- The Developer Dilemma: Adler highlights a fundamental issue facing frontier labs: even the engineers and researchers building these advanced AI systems do not fully understand what boundaries the models will reliably respect or obey once deployed.
- The Competitive Trap: Leading AI firms continue to accelerate developmentdespite acknowledging catastrophic or existential riskslargely driven by the race dynamics and the fear that competitors or rival nations will build them first.
- Call for Government Intervention: Because competitive pressures prevent companies from slowing down voluntarily, Adler argues that government regulation is essential to establish clear, enforceable safety standards. He also points to international diplomacy, urging major powers like the United States and China to establish bilateral treaties and global safety guardrails.
- Immediate Corporate Responsibility: In addition to state regulation, Adler argues that major AI companies must immediately adopt stricter internal safety mechanisms, respect their own risk thresholds, and prevent dangerous models from operating unchecked. Failing to do so, he warns, will erode public and employee trust and could eventually lead to outright bans on AI technology.