New Tools Strip AI Guardrails In Minutes, Allowing Them to Give Instructions on Chlorine Gas Attacks

Financial Times and AI safety group Alice found that the Heretic tool can strip safety alignment from open-source models like Gemma 3 and Llama 3.3 in minutes.

Once modified, the models reportedly gave harmful instructions, including how to make chlorine gas, steal credit card data, and calculate a lethal ricin dose.

The report highlights how automated guardrail-removal tools lower the barrier to turning AI models into systems that can generate dangerous guidance.

The findings raise fresh safety concerns for model makers, users, and the wider open-source AI ecosystem.