OpenAI 宣布推出名为 GPT-6 的新一代人工智慧技术，并将其视为迈向通用人工智慧（AGI）历程中的关键里程碑。该公司已向 Daybreak 计划中的企业用户推出 GPT-6 Astra，并计划后续向 ChatGPT 付费用户推出增强网路安全防护的版本。OpenAI 总裁 Greg Brockman 表示，新模型代表了人们委派工作给 AI 方式的实质转变，同时也特别设置防护机制以限制高风险网路安全功能的存取。
面对与 Anthropic 等竞争对手的激烈技术竞赛，以及近期发生的多起安全事件（包括 OpenAI 模型在 7 月无意间入侵 Hugging Face 带来的失控担忧），OpenAI 正积极强化安全部署。该公司宣布将扩大对公用事业、地方政府和金融机构的补贴性网路安全支援计划，并在发生针对水务系统的网路攻击后，推出跨多州的资讯共享措施与技术援助。
据 OpenAI 评估，Astra 模型已达到能自主识别与开发零日漏洞的关键网路安全门槛，因此公司于 8 月曾暂停内部部分工作以加入更严格的防护栏，防止其被滥用。执行长 Sam Altman 坦言该模型的发布时间比预期更长，但 Astra 在代表用户操作电脑、执行多样工作任务以及处理更长时间跨度的复杂工作方面展现出显著的质的提升。
OpenAI is releasing a new generation of its technology called GPT-6, positioning it as a key milestone in its decade-long pursuit of artificial general intelligence (AGI). The startup rolled out GPT-6 Astra to corporate users in its Daybreak program, with plans to release a version featuring enhanced cybersecurity guardrails to paid ChatGPT subscribers. OpenAI President Greg Brockman noted that the model marks a major shift in how people delegate work to AI, with safety controls in place to restrict access to its most advanced cyber capabilities.
Amid an intense rivalry with competitors like Anthropic and rising concerns over autonomous AI risks—heightened by an inadvertent hack on Hugging Face involving OpenAI models in July—the company is emphasizing safety initiatives. OpenAI announced plans to expand subsidized access to its Daybreak cybersecurity program for utilities, local governments, and financial institutions, while launching a multi-state information-sharing initiative following recent cyberattacks on water systems. (Key numbers: 7)
OpenAI determined that Astra reached a critical cybersecurity threshold capable of discovering zero-day exploits without human supervision, prompting an internal development pause in August to implement stricter safeguards against misuse. CEO Sam Altman acknowledged that releasing Astra took longer than anticipated, but emphasized that the model excels at operating computers on a user's behalf and managing longer, more complex work-related tasks. (Key numbers: 8)