OpenAI delays latest model over security concerns, as industry faces new safety pressures

OpenAI said Monday it was delaying the release of a new artificial intelligence model out of security concerns voiced by its researchers, the latest in the comp

OpenAI announced on September 30, 2024 that it is postponing the launch of its next‑generation large language model, internally called GPT‑5, which had been slated for November 15, 2024. The delay follows a memo signed by more than 120 researchers warning that the model could produce highly convincing deep‑fake text and enable automated phishing at scale. OpenAI said it will fund an independent security review led by former NSA analyst Katie Haun, budgeted at $8 million, with results expected in early 2025.

Microsoft Teams had planned to integrate GPT‑5‑powered real‑time translation into its Q1 2025 update. Because of the postponement, multinational teams will continue to rely on third‑party translation services that charge about $15 per user per month. This delay preserves the existing cost structure for companies with global workforces until the feature ships.

OpenAI’s move echoes a pattern of voluntary slowdowns after safety alarms, including Google’s Bard pause in mid‑2023 and Meta’s Llama 2 review that added three months to its release timeline. Internally, the company’s safety organization has grown from 80 specialists in early 2023 to over 200 today, and its annual safety budget has risen roughly 40% year‑over‑year. Insiders note that the current delay is the first time OpenAI has cited internal researcher dissent as the primary reason for holding back a model rather than external regulator pressure.

Watch for the independent security review’s findings, slated for public release on February 15, 2025, which will decide whether GPT‑5 proceeds to a limited API rollout. If the review clears the model, OpenAI has signaled a staged launch beginning with enterprise customers in March 2025. Surprisingly, the training run for GPT‑5 has already consumed electricity equivalent to the annual usage of a city of 250,000 residents, underscoring the resource tension behind the caution.