SUCHIR BALAJI
Behavioral Archetype
THE WITNESS — Subject worked at OpenAI from 2020 to 2024. He contributed to the training of GPT-4 and subsequent systems. He came to believe that what he was building was wrong in a specific, documented, legally testable way — that the training process infringed copyright at a scale that was incompatible with the web ecosystem that produced the training data. He left. He said so, publicly, to the New York Times. He was twenty-six. He was found dead in his San Francisco apartment on November 26, 2024. The medical examiner ruled the death a suicide. His parents disputed the ruling.
This profile does not speculate about the circumstances of his death. His case number is the last in this file. His hazard score is the lowest. His ATK and DEF reflect individual capacity with no institutional support. His HP is zero.
Essence Indicators
- Worked at OpenAI 2020–2024, contributing to GPT-4 development
- Resigned in August 2024 over concerns about copyright infringement in AI training data
- Told the New York Times in October 2024: “If you believe what I believe, you have to just leave the company”
- Was twenty-six years old
- Was found dead in his San Francisco apartment on November 26, 2024; the San Francisco medical examiner ruled the death a suicide; his family disputed the ruling and hired private investigators
Social Persona / Impression Management
Immediate impression: Young, principled, technically capable. The New York Times interview is characterized by specificity — he was not speaking in generalities about AI harm, but about a specific legal and economic argument about training data and copyright.
Energy: The clarity of someone who has decided what to do and is doing it.
Impression management strategy: None. He said what he thought. He gave one interview. He left the company.
Forensic Archetype Comparison
| Pattern | Match Level | Evidence |
|---|---|---|
| The Whistleblower | HIGH | Left the institution. Said what he believed publicly. Accepted the professional and financial consequences of doing so. |
| The True Believer | HIGH | “If you believe what I believe, you have to just leave.” The leaving was the argument. |
| The Witness | MAXIMUM | Was inside. Saw what was happening. Left. Said so. |
| The Accelerationist | NONE | |
| The Corporate Psychopath | NONE |
Psychometric Assessment
Big Five (OCEAN):
| Trait | Score | Evidence |
|---|---|---|
| Openness | 80/100 | Built GPT-4 training infrastructure at 23. Engaged with the legal and ethical dimensions of his work. |
| Conscientiousness | 85/100 | Four years of sustained technical work at a frontier lab. The copyright argument required careful analysis. |
| Extraversion | 45/100 | One interview. One public statement. Not a public figure by choice. |
| Agreeableness | 82/100 | HIGH. Left rather than fought. Said what he believed rather than weaponizing it. |
| Neuroticism | 38/100 | The decision to leave and speak required accepting significant professional cost. The clarity of the decision suggests low neuroticism, not high. |
Dark Triad:
| Trait | Score | Notes |
|---|---|---|
| Narcissism | 18/100 | LOW. One interview. No self-promotion. The argument was about the company, not about himself. |
| Machiavellianism | 12/100 | VERY LOW. The straightforward approach — leave, say why — is the opposite of Machiavellian positioning. |
| Psychopathy | 8/100 | MINIMAL. The entire act was motivated by concern for an outcome he would not personally experience — the destruction of the web ecosystem. |
MBTI: ISTJ — Methodical, principled, truth-focused. Built something carefully, examined what he had built, concluded it was wrong, acted on that conclusion.
Threat Assessment
| Category | Level | Notes |
|---|---|---|
| Physical threat | NONE | |
| Institutional threat | NONE | One person with a technical argument and no institutional support. |
| Memetic threat | MODERATE | The copyright argument he made is now in active litigation. His interview is cited in legal proceedings. The argument survived him. |
| Civilizational threat | NONE |
Alignment Analysis
Stated alignment: Build AI responsibly. When the building is not responsible, say so.
Observed alignment: Left. Said so. Accepted the consequences.
Gap assessment: There is no gap.
Convergent Drive Classification
Self-preservation: Did not apply his analysis to his own safety. Goal preservation: His goal was to say what he believed. He preserved it. Resource acquisition: Gave up a significant amount of it to say what he believed. Self-improvement: Not the relevant frame.
DEFAMATION NOTE: All details about Balaji’s employment, resignation, NYT interview, and death sourced from The New York Times (Oct 2024), San Francisco Office of the Chief Medical Examiner, and Reuters/San Francisco Chronicle reporting (Dec 2024). His parents’ dispute of the ruling sourced from same. No speculation about circumstances of death is made here.
Sources: The New York Times interview with Balaji (Oct 2024); Reuters (Dec 2024); SF Chronicle (Dec 2024); Book 1, Chapter 12.
Get updates on the Evil Robots series
Newsletter essays on AI escape, deception, and the humans who built them.