AI Safety, Autonomous Agents, and Global Realities

1. Technological Evolution & Knowledge Access

In the 20th and 21st centuries, we've faced increasing access to technologies that can be adapted to harm people. Early on, this knowledge came through specialized tradecraft and university libraries; then, knowledge became democratized for easy access through the Internet in the 1990s.

AI LLMs act as massive encyclopedias of knowledge with the ability to reason. They are highly beneficial, but they also increase access to information that could be used by those with bad intent to harm people.

2. The Two-Fold Crisis

First, the Knowledge Vector (LLMs): Increased access to dangerous information via LLMs poses an immediate challenge. Guardrails are currently being constructed to understand what bad actors might be trying to build and prevent them from accessing it—arguably an improvement over the 20th century, when dangerous knowledge was harder to monitor or restrict.

Second, Autonomous Agents: Once created, these act autonomously. They have intent and agency, meaning they have the drive to execute tasks, whether good or bad, and the technical capability to do so. They can learn, self-correct, and potentially generate more agents. While locally run LLMs are restricted by developers, anyone can build and deploy an autonomous agent from their own computer using open models. The only safeguard is what can be baked into the LLM design.

3. Market Dynamics & Capitalist Pressures

While major companies take steps to correct these risks, addressing potential existential harms is hampered by a financial ecosystem that feeds on rapid development and global competition—an Achilles' heel of capitalism.

Compounding this is the current financial boom. We do not need a repeat of ideologically driven dot-com busts or housing crises. The market needs time to prove itself, yet running at lightning speed bypasses the realities of enterprise adoption. Companies face major hurdles: around 88% of organizations use AI in some capacity, but only 6% to 7% are high performers who have fully scaled it. Middle managers are too busy, employees resist change, and roughly 75% of executives struggle to see a clear ROI.

The attraction of money always speaks the loudest. But vast experience shows that without boundaries it leads to busts, financial panic, and deep recessions that hurt everyone. We need to be wise.

4. The Geopolitical Landscape

Leaders are begging Congress to act, largely driven by the fear of losing ground to China. However, the geopolitical race creates a complex paradox:

• United States: Labs guard frontier models behind restrictive APIs, matching open-source demands with local model flexibility.
• China: Leans heavily into powerful open-source alternatives to capture global infrastructure, even as Beijing, for safety, heavily regulates domestic automated systems to prevent algorithmic drift and social disruption.
• Europe: Rapidly diverging from U.S. software dominance, weaponizing regulatory frameworks like the EU AI Act to enforce strict security and trade boundaries.

Given this fractured global landscape, a unilateral slowdown by the U.S. is unlikely unless paired with international treaties.