AI Responsibility – OpenAI and Anthropic

TEXT START: I resigned from Anthropic today.

1. The Dissection

This is an insider alarm flare, not a demonstrated case. The author uses three years of pretraining work at both OpenAI and Anthropic as an authority credential, then makes two sweeping claims: neither company is acting responsibly, and both are racing toward self-improving superintelligence. The supplied text gives no decisions, evidence, technical milestones, or mechanism. Its function is to convert private dissent into public pressure.

2. The Core Fallacy

The post frames the danger primarily as a failure of corporate responsibility. Under the Discontinuity Thesis, that is too shallow. If self-improving AI creates decisive strategic and economic advantage, competition makes acceleration the default; conscience is not a durable brake. A company that unilaterally slows may surrender the asset to a rival or state. The warning may be true, but its moral diagnosis mistakes the visible actors for the underlying machine. It also presents the trajectory as fact without proving it.

3. Hidden Assumptions

• Self-improving superintelligence is technically attainable and near enough to drive present behavior.
• Safety can be made decisive against competitive and state incentives.
• Public disclosure or resignation can force coordination.
• Institutions can preserve meaningful human control at scale, despite P2: Coordination Impossibility.
• “Safety” is a single objective; the post does not distinguish technical alignment, deployment control, ownership, labor displacement, or distribution of power.
• The consequences fall on “our lives” generally, rather than concentrating power and income in the owners of AI capital.

4. Social Function

Classification: partial truth, amplified by prestige signaling, with an elite self-exoneration function. The insider credential makes the warning harder to dismiss. The resignation provides a moral exit for the dissenter, while the focus on corporate conduct can obscure the deeper ownership-and-competition structure. It is not copium. It is an alarm that stops before naming the furnace.

5. The Verdict

The post is directionally serious but evidentially thin and structurally incomplete. It identifies acceleration toward systems that could make P1—Cognitive Automation Dominance—real, but it does not trace the consequence through P2 and P3: human institutions fail to preserve human-only economic domains, and productive participation collapses. If the thesis holds, the labs do not need to be uniquely reckless for the race to continue; the race itself is the kill mechanism. Responsibility rhetoric without control of the capital, compute, energy, and deployment channels is ceremonial language spoken beside a loaded weapon.