If the court decides that ChatGPT becomes open source, Dusoma Foundation recommends a 60-day safety evaluation before the source code is released. This will allow us to ensure that someone can’t immediately use it to release a cyber attack or bio weapon.
The 60-Day Global Audit
If the “weights” of models like GPT-4 are forced into the wild, we shouldn’t just hit “upload” and hope for the best. Instead, we are proposing a mandatory 60-day pause.
During this window, we enlist a “Global 10,000″—the top minds in cybersecurity, ethics, and neural architecture—to perform a decentralized, high-pressure audit of the model. This isn’t a private corporate review; it’s a public safety inspection.
A New Framework for Safety
The goal of this 60-day breath is to develop a safety plan for sanitizing the model before it hits the general public. We need to move beyond simple filters and look at more permanent technical solutions:
- Machine Unlearning: Developing protocols to effectively “scrub” dangerous, high-risk knowledge (like chemical weapon precursors) from the model’s neural weights.
- Adversarial Noise: Feeding the model conflicting data on sensitive topics to ensure it cannot provide actionable, harmful instructions.
- The “Middle Path”: Moving away from the binary choice of “Corporate Monopoly” vs. “The Wild West.”
Why This Matters Now
The testimony coming out of the Ronald V. Dellums Federal Building suggests that OpenAI’s defense relies heavily on the “bioweapon” fear—using it as a shield to justify a for-profit pivot.
By implementing a staged, audited release, we take that excuse off the table. We prove that transparency and safety aren’t mutually exclusive; they just require a bit of collective patience.
The public was promised open technology. Let’s make sure it’s safe enough to actually own.
Kathryn Born, Editor, Starve Magazine