Former Anthropic Researcher Warns AI Labs Are 'Gambling With Our Lives'

Jacob Coxon, a veteran AI researcher, has resigned from Anthropic and issued a stark warning that the industry's pursuit of superintelligence is reckless. The resignation has immediately shifted the conversation in Washington, where lawmakers are now questioning the safety protocols of the world's leading AI labs.

Coxon, who spent three years in pretraining research at both OpenAI and Anthropic, took to X (formerly Twitter) to voice his concerns. He claimed that the leading companies in the field are "racing straight to self-improving superintelligence and gambling with our lives." According to Coxon, neither OpenAI nor Anthropic is acting responsibly as they push toward the creation of systems capable of autonomous improvement.

The Safety Debate

This departure comes during a period of intense scrutiny regarding the trajectory of frontier AI development. The tension within the industry is further highlighted by the reactions of other insiders. Following Coxon's resignation, Evan Hubinger, an alignment lead at Anthropic, echoed the sense of urgency in a post stating there is a greater than 10% chance that AI could kill all humans.

These warnings are particularly significant because they originate from the pretraining level—the foundational stage of model creation. This indicates that the concerns are rooted in the core architecture of how these systems are built, rather than just their final applications. The rift between the commercial drive for capability and the internal demands for safety is becoming increasingly apparent as researchers leave the most prominent labs.

Political Fallout

Coxon's insider perspective has amplified calls for government intervention in the United States. The warning caught the attention of several U.S. lawmakers who view the potential for uncontrolled superintelligence as a national security risk. Senator Ted Cruz reacted strongly to the researcher's claims, describing the situation as "scary as hell."

As Washington reacts, the focus is expected to shift toward whether voluntary safety commitments from AI labs are sufficient or if mandatory federal oversight is required. While the political reaction has been swift, it remains to be seen if these warnings will lead to concrete legislative action or a shift in how companies like OpenAI and Anthropic manage their development timelines.

The current trajectory suggests a collision course between rapid innovation and existential risk. If the foundational researchers responsible for building these models believe the process is reckless, the pressure for a formal regulatory framework will likely intensify, moving beyond the current system of voluntary guidelines.