Ilya Sutskever left OpenAI on May 14, ending a nearly decade-long tenure that included the controversial boardroom coup against Sam Altman. He was followed immediately by Jan Leike, a safety researcher who co-led the Superalignment team with him. The exit of these two figures forced the immediate dissolution of the group tasked with ensuring superintelligent AI remains safe. OpenAI had pledged 20% of its computing resources to this initiative over four years, but Leike stated that the company failed to deliver on that promise. Instead, he argued that "safety culture and processes have taken a backseat to shiny products." This breakdown in trust signaled a fundamental shift in how one of the world’s top AI labs prioritizes risk versus speed.
A straight shot to secure intelligence
Sutskever did not leave the field; he founded Safe Superintelligence Inc., or SSI, with Daniel Gross and Daniel Levy. The company operates as a for-profit entity headquartered in Palo Alto with offices in Tel Aviv. It raised $1 billion in a Series A round at a $5 billion valuation, backed by Andreessen Horowitz, Sequoia Capital, DST Global, and SV Angel. Investors are making outsized bets on exceptional talent despite a general waning of interest in unprofitable foundational research firms. SSI describes its approach as a "straight shot" to safe superintelligence, focusing exclusively on one goal and one product. The team plans to use the funds to acquire computing power and hire a small, highly trusted group of researchers. Gross emphasized that they spend hours vetting candidates for "good character" rather than just credentials. They seek people interested in the work, not the hype. This concentrated strategy contrasts sharply with the broader, more diffuse efforts at larger tech giants.
The financial backing underscores a specific belief: superintelligence is within reach and building it safely is the most important technical problem of our time. Sutskever noted that he identified a mountain different from what he was working on at OpenAI. He plans to approach scaling differently, asking "what are we scaling?" rather than assuming performance improves automatically with compute. This philosophical pivot suggests that raw horsepower alone will not solve alignment issues. The company aims to partner with cloud providers and chip companies to fund infrastructure needs, though specific partners remain undecided. By isolating the mission, SSI avoids the internal conflicts that plagued OpenAI’s Superalignment team. There, core members like Leopold Aschenbrenner and Pavel Izmailov were fired for leaking information, while others left due to loss of confidence. The disbandment left OpenAI with less preparation for AGI, a gap Sutskever now intends to fill.
The gap between promise and practice
OpenAI’s decision to dismantle its dedicated safety team raises serious questions about the future of AI safety at a leading lab. John Schulman, OpenAI’s new scientific lead, appeared to dodge tough questions about readiness for AGI during an interview on The Dwarkesh Podcast. When asked what the plan is if AGI arrives sooner than expected, Schulman seemed to want the interview to end. Paul Roetzer, founder of Marketing AI Institute, summarized the response bluntly: "Basically, he has no idea. No plan." This admission highlights a critical vulnerability. If superintelligence happens, OpenAI is now less prepared for it without the specialized focus of the dissolved team.
Leike’s final warning encouraged employees to "feel the AGI," a phrase used internally to recognize that there is no obvious human-level ceiling for AI capabilities. It is a call to take moral obligations seriously. Not everyone agrees on the urgency. Yann LeCun at Meta vocally disagrees that humanity has even figured out how to design a superintelligent system, arguing fears are severely overblown. However, the divergence in opinion does not negate the risk. SSI’s existence validates the concern that safety requires dedicated resources and clear leadership. The industry is split on whether caution slows innovation or prevents catastrophe. What is safe, what is alignment with human values, and how do you stay competitive while balancing that with safety? Maybe someday it will be very clear what is safe and what isn’t. I am not at all convinced that right now anyone has that answer.
