OpenAI has released preliminary cybersecurity evaluations for its Astra model, highlighting potential vulnerabilities. The company is also outlining measures to improve safety and security features. These evaluations are part of OpenAI's efforts to make AI more transparent and accountable. --- Why it matters: This matters because it shows how AI developers are acknowledging and addressing the risks associated with their models. Engineers can learn from these evaluations and apply similar approaches to strengthen their own AI systems' defenses. Source: https://openai.com/index/responding-next-frontier-critical-cyber-capabilities