光速滑跪之后，OpenAI公布与五角大楼协议，但引来更多质疑

Core Viewpoint - OpenAI has reached an agreement with the Pentagon to deploy its AI models in a secure environment, emphasizing that the contract includes strict safeguards to prevent misuse, contrasting with Anthropic's refusal to comply with similar demands from the Department of Defense [1][2][3]. Summary by Sections OpenAI's Agreement with the Pentagon - OpenAI announced a contract with the Pentagon to deploy AI models in a classified network, claiming that this agreement includes three core "red lines" that restrict the use of its technology [4][6]. - The three red lines are: 1. The technology cannot be used for large-scale domestic surveillance 2. It cannot be used to command autonomous weapon systems 3. It cannot be used for high-risk automated decision-making systems [7][10]. Comparison with Anthropic - Anthropic faced backlash for refusing to grant the Pentagon full access to its AI systems, leading to its exclusion from government contracts, while OpenAI's agreement has been met with skepticism regarding its ethical implications [1][3][16]. - OpenAI's safeguards are claimed to be more robust than those in Anthropic's previous agreements, focusing on cloud-only deployment and continuous oversight by qualified personnel [16][19]. Deployment Architecture and Protocols - All deployments will be cloud-based, ensuring that OpenAI maintains control over the security measures and compliance with the established red lines [8][10]. - The agreement stipulates that the Pentagon can use the AI system for lawful purposes, but it must adhere to existing laws and regulations regarding human oversight in high-stakes decisions [10][11]. Expert Involvement and Oversight - OpenAI will provide qualified engineers and researchers to support the Pentagon, ensuring that the deployment aligns with the agreed-upon safety protocols [13][15]. - The company emphasizes that it will not compromise its core protective measures for the sake of enhancing national security capabilities [15][20]. Public Concerns and Criticism - Despite OpenAI's assurances, there are concerns about the vagueness of terms like "all lawful purposes," which could potentially allow for broader interpretations that may lead to misuse [26][28]. - Critics argue that the safeguards may not be sufficient to prevent future violations of the red lines, suggesting that the agreement could be a "blank check" for the Pentagon [28][30][37].