What Are the 5 Big AGI Readiness Gaps Ex-OpenAI Researcher Miles Brundage Says We Are Not Ready For?

TL;DR

Neither OpenAI, other frontier labs, nor the wider world is ready for AGI, according to former OpenAI policy researcher and senior advisor for AGI readiness Miles Brundage. He identifies five gaps: shared understanding, regulatory infrastructure, legitimacy, societal resilience, and differential development. His October 2024 departure also gives him greater independence to address these problems, making the reasoning behind each gap worth examining.

Transcript

Yet another safety researcher at OpenAI has left. Miles Brundage, who was the head of policy research at OpenAI for no less than six years, moving into the role of senior advisor for AGI readiness. Publicly left OpenAI this October 2024, and he wrote an entire Substack article about it. Now, there are a lot of people that have left OpenAI. It seems... Read More

Key Insights

• Independence changes policy credibility: Brundage did not present his departure solely as dissatisfaction with OpenAI. He said he had achieved much of what he set out to do, while also seeking freedom over his public work. Independence could reduce both internal publishing constraints and the public perception that his policy positions primarily reflect the interests of a major AI company.
• Senior departures amplify concern: The presenter treats Brundage's exit as especially significant because he spent six years leading policy research before becoming senior advisor for AGI readiness. His departure is situated within a broader pattern involving research and safety personnel. The concern is not ordinary turnover alone, but whether specialists believe they can conduct their most beneficial work from inside OpenAI.
• Safety work moved elsewhere: Ilya Sutskever is presented as another prominent example of someone leaving OpenAI and creating an independent organization. His company, Safe Superintelligence, is described as having raised $1 billion around a single mission involving safe artificial intelligence systems in the AGI domain. The example supports the presenter's wider focus on safety expertise moving beyond established frontier laboratories.
• The timeline shapes urgency: The presenter argues that AGI may arrive by 2029, if not sooner, rather than remaining centuries or decades away. That forecast drives the demand for immediate preparation across policy, economics, and technical safety. Within the video's argument, readiness work cannot wait for universal agreement because rapidly improving capabilities could outrun the institutions expected to govern their consequences.
• Shared understanding comes first: Brundage sees a major knowledge deficit among technology companies, governments, and the public. These groups do not share a reliable picture of what AGI is or what its implications could be. Without that foundation, public debate, policymaking, and participation can proceed from incompatible assumptions, weakening every other attempt to prepare for increasingly capable systems.
• Practical familiarity supports literacy: The existing material presents direct experience with current AI tools as an accessible starting point for individuals. Using these systems in everyday work and exploring appropriate automation can reveal both their capabilities and implications. This practical literacy does not solve AGI governance, but it can improve the quality of conversations that otherwise remain driven by abstraction, confusion, or fear.
• Legacy rules face new systems: Regulatory infrastructure is considered inadequate because current laws and policies were not created with AGI in mind. The problem is not merely the absence of one new rule. It is a mismatch between rapidly advancing, broadly consequential technology and governance frameworks designed for much simpler systems, which is why Brundage calls for greater urgency from policymakers.
• Participation depends on knowledge: Legitimacy requires consequential AI decisions to move beyond closed rooms and receive wider public involvement. Yet participation becomes less meaningful when people lack even a basic understanding of present AI capabilities. The material therefore links legitimacy to shared understanding: education improves people's ability to evaluate risks and choices, while openness gives their informed judgments a place in decision-making.
• Resilience includes economic redesign: AGI readiness extends beyond preventing technical failures. Society may need safeguards and new economic arrangements if labor becomes detached from capital or conventional paid work no longer performs its familiar role. Discussing those possibilities before disruption arrives is part of resilience because institutions will otherwise be forced to respond only after economic and social pressures have intensified.
• Control must match capability: Differential development describes the gap between rapid gains in what AI systems can do and slower gains in controlling those systems. The rocket-engine comparison makes the imbalance concrete: greater power is not sufficient when steering remains unresolved. Readiness therefore requires safety, control, and governance mechanisms to progress with capability development, not appear only after powerful systems exist.
• Risk and opportunity coexist: The presenter does not frame AGI solely as catastrophe. Alongside extreme dangers, the video describes the possibility of greater health, abundance, and resources if humanity adapts, plans new economic paradigms, and stewards the technology well. This dual framing makes readiness both defensive and constructive, concerned with avoiding harmful outcomes while creating conditions for broadly beneficial ones.
• Responsibility operates at several levels: Individuals can increase their understanding and discuss AI more openly, but the largest readiness gaps cannot be closed through personal adaptation alone. Laboratories influence capability and control development, policymakers shape regulatory infrastructure, and public participation affects legitimacy. Brundage's framework consequently distributes responsibility while emphasizing that each group depends on progress by the others.

Questions & Answers

Q: What are the five biggest gaps in AGI readiness?

Miles Brundage identifies shared understanding, regulatory infrastructure, legitimacy, societal resilience, and differential development as the five biggest gaps. Together, they cover whether people understand AGI, whether suitable rules exist, whether decisions receive legitimate public input, whether society can absorb disruption, and whether control methods keep pace with capabilities. The framework matters because progress in AI capability alone does not create readiness. Brundage's overall judgment is that OpenAI, other frontier laboratories, and the world have not yet closed these gaps.

Q: Why did Miles Brundage leave OpenAI in October 2024?

Brundage said he had completed much of what he originally set out to accomplish during six years at OpenAI. He also wanted more autonomy over what he published and said, which independence could provide beyond company red tape and editorial constraints. He believed audiences may perceive policy ideas from an OpenAI employee as influenced by the company. Leaving therefore gave him a chance to pursue urgent AI policy work with greater freedom and potentially greater public credibility.

Q: Why is shared understanding essential for AGI readiness?

Shared understanding is essential because technology companies, government officials, and the public lack a common grasp of AGI and its implications. Without a common foundation, these groups can debate policy or participate in decisions while working from conflicting assumptions. Familiarity with current AI tools is presented as one practical way for individuals to improve their understanding of capabilities and automation. Better understanding also supports the legitimacy gap because informed participants can evaluate risks, opportunities, and policy choices more meaningfully.

Q: Why is current regulatory infrastructure inadequate for AGI?

Existing laws and policies were not designed for systems with AGI's potential breadth and power. As AI capabilities improve rapidly, rules created for simpler technologies may fail to address the decisions and consequences associated with advanced systems. Brundage therefore argues that policymakers must act with more urgency rather than waiting for capability development to slow down. Modernized regulatory infrastructure is needed because governance must be capable of responding while the technology is advancing, not only afterward.

Q: How can AGI decision-making become more legitimate?

Greater legitimacy requires opening consequential AI decisions beyond closed rooms and creating meaningful opportunities for public involvement. Participation alone is insufficient, however, when much of the public lacks a basic understanding of current AI tools or AGI's possible implications. Education and transparency must therefore develop together so people can contribute with relevant context. This combination matters because decisions carrying broad social consequences need both informed scrutiny and participation beyond the laboratories building the systems.

Q: What does societal resilience mean for AGI preparation?

Societal resilience means developing safeguards, institutions, and economic arrangements that can absorb major social and economic disruption. The material considers a future in which labor may become detached from capital and conventional work may no longer be necessary for meeting basic expenses. Preparing involves discussing new economic models before those changes arrive rather than improvising only after disruption. This gap matters because technical control of AGI would not by itself resolve the consequences for employment, income, or social organization.

Q: What is differential development in artificial intelligence?

Differential development is the imbalance between rapidly improving AI capabilities and the slower development of ways to control them. The material compares this to building a rocket engine without knowing how to steer, separating raw power from reliable direction. Closing the gap requires safety, control, and governance mechanisms to develop alongside increasingly capable systems. The reason is straightforward: capability growth can increase consequences faster than society's ability to guide or constrain those capabilities.

Q: What can individuals do now to become more prepared for AGI?

Individuals can start by using current AI tools, examining appropriate automation in their work, and building practical awareness of what the systems can and cannot do. They can also discuss AI's implications openly instead of approaching the subject only through fear or abstraction. These activities help close the shared-understanding gap and create a stronger foundation for informed public participation. They do not replace the responsibilities of laboratories and policymakers, but they improve the public knowledge needed for legitimate governance and broader preparation.

Summary & Key Takeaways

• A notable OpenAI departure: Miles Brundage publicly left OpenAI in October 2024 after six years as head of policy research and later senior advisor for AGI readiness. The presenter places his exit alongside departures by Ilya Sutskever and others associated with research and safety. Sutskever subsequently started Safe Superintelligence, described as having raised $1 billion for a mission focused on developing safe artificial intelligence systems in the realm of AGI.

• Why Brundage chose independence: Brundage said he had completed much of what he originally intended to accomplish, making part of his departure a natural career decision. He also wanted greater autonomy over what he could publish and say. The presenter notes that editorial rules and organizational red tape can constrain people inside large technology companies. Brundage additionally believed that audiences may question policy proposals originating within the AI industry because employment at OpenAI creates a perception of bias.

• The readiness verdict is stark: Brundage asked how OpenAI and the wider world were performing on AGI readiness. His answer was that neither OpenAI nor any frontier laboratory was ready, and the world was not ready either. He argued that AI capabilities were improving very rapidly while policymakers were not responding with enough urgency. Independence, in his view, could let him contribute more effectively to policy work intended to close that growing readiness deficit.

• Five gaps define the problem: Brundage identified shared understanding, regulatory infrastructure, legitimacy, societal resilience, and differential development as the central deficiencies. Stakeholders lack a common grasp of AGI, existing rules were not designed for systems of this potential power, and important decisions often lack broad participation. Society is also insufficiently prepared for economic disruption, while methods for controlling advanced systems risk progressing more slowly than the systems' underlying capabilities.

• Preparation requires coordinated action: The proposed response spans education, governance, participation, economic planning, and technical control. Individuals can build familiarity with current AI tools and discuss their implications, while policymakers and laboratories carry responsibility for modernizing rules and opening consequential decisions to legitimate scrutiny. Society must also consider safeguards and new economic arrangements if conventional work changes substantially. Above all, safety and control capabilities need to advance alongside raw AI capabilities rather than following after deployment.