OpenAI vs Anthropic: The Results of the AI Safety Test - AI Magazine
The evaluation reveals that Claude models are most vulnerable to “past tense” jailbreaks, where harmful requests are reframed as historical questions.
Read the original at AI Magazine ↗
- Topics
- AI Safety & Existential RiskCybersecurityMarkets & Investment
- Mentions
- AnthropicClaudeOpenAI
- Tracked via
- “dangerous AI”
Same story, other outlets
- What Can We Learn from OpenAI & Anthropic's AI Safety Test? | Technology Magazine Technology Magazine · Sep 01, 2025
- OpenAI vs Anthropic: The Results of the AI Safety Test - AI Magazine AI Magazine · Sep 02, 2025