Safety Research Grants

• Up to $50,000 in credits for safety research on Thinking Machine's open-weight models.
• Directions include: differential defensive capabilities, hazardous data filtering, and tamper-resistant training.
• Investigate alignment failure modes from fine-tuning, reward hacking, and oversight gaming.
• Estimate worst-case and marginal risk. Forecast safety-relevant scaling trends.
• Use adversarial fine-tuning to test whether safeguards persist or merely suppress.

• Up to $50,000 in credits for safety research on Thinking Machine's open-weight models.
• Directions include: differential defensive capabilities, hazardous data filtering, and tamper-resistant training.
• Investigate alignment failure modes from fine-tuning, reward hacking, and oversight gaming.
• Estimate worst-case and marginal risk. Forecast safety-relevant scaling trends.
• Use adversarial fine-tuning to test whether safeguards persist or merely suppress.

Thinking Machines is an AI startup founded by Miri Murati.