Cinder, a platform for stopping AI-powered abuse, uses a technique called "model obliteration." This involves intentionally removing the built-in safety guardrails from open-source models. By doing so, they can train the AI on harmful content and create more effective, specialized classifiers to detect abuse at scale.

Simple refusal mechanisms in AI models are easily bypassed by motivated actors. Effective biosecurity requires deeper interventions, such as curating training data to exclude sensitive biological information or implementing strict access controls for the most powerful models, ensuring they aren't publicly available.

The open-source model ecosystem enables a community dedicated to removing safety features. A simple search for 'uncensored' on platforms like Hugging Face reveals thousands of models that have been intentionally fine-tuned to generate harmful content, creating a significant challenge for risk mitigation efforts.

China remains committed to open-weight models, seeing them as beneficial for innovation. Its primary safety strategy is to remove hazardous knowledge (e.g., bioweapons information) from the training data itself. This makes the public model inherently safer, rather than relying solely on post-training refusal mechanisms that can be circumvented.

The dangerous side effects of fine-tuning on adverse data can be mitigated by providing a benign context. Telling the model it's creating vulnerable code 'for training purposes' allows it to perform the task without altering its core character into a generally 'evil' mode.

A novel safety technique, 'machine unlearning,' goes beyond simple refusal prompts by training a model to actively 'forget' or suppress knowledge on illicit topics. When encountering these topics, the model's internal representations are fuzzed, effectively making it 'stupid' on command for specific domains.

Instead of trying to control open-source AI models, which is intractable, the proposed strategy is to control the small, expensive-to-produce functional datasets they train on. This preserves the beneficial open-source ecosystem while preventing the dissemination of dangerous capabilities like viral design.

Research on bio-foundation models like EVO2 and ESM3 shows that strategically excluding key datasets (e.g., sequences of viruses that infect humans) dramatically reduces a model's performance on dangerous tasks, often to random chance, without harming its useful scientific capabilities.

A novel AI safety technique called gradient routing trains mixture-of-experts models to isolate dangerous knowledge (e.g., bioweapons, cyber exploits) into specific "expert" modules during pre-training. These dangerous experts can then be completely removed ("ablated") before deployment, creating an inherently safer model.